modelId
string | author
string | last_modified
timestamp[us, tz=UTC] | downloads
int64 | likes
int64 | library_name
string | tags
list | pipeline_tag
string | createdAt
timestamp[us, tz=UTC] | card
string |
|---|---|---|---|---|---|---|---|---|---|
PrunaAI/resnet152.a1_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:21:25Z | 2 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T09:20:56Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir resnet152.a1_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/resnet152.a1_in1k-turbo-green-smashed --local-dir resnet152.a1_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "resnet152.a1_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "resnet152.a1_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model resnet152.a1_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/resnetv2_50d_gn.ah_in1k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:21:24Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T09:47:44Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir resnetv2_50d_gn.ah_in1k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/resnetv2_50d_gn.ah_in1k-turbo-tiny-green-smashed --local-dir resnetv2_50d_gn.ah_in1k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "resnetv2_50d_gn.ah_in1k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "resnetv2_50d_gn.ah_in1k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model resnetv2_50d_gn.ah_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/convmixer_1024_20_ks9_p14.in1k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:21:23Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-07T16:21:28Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir convmixer_1024_20_ks9_p14.in1k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/convmixer_1024_20_ks9_p14.in1k-turbo-tiny-green-smashed --local-dir convmixer_1024_20_ks9_p14.in1k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "convmixer_1024_20_ks9_p14.in1k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "convmixer_1024_20_ks9_p14.in1k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model convmixer_1024_20_ks9_p14.in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/ecaresnet50t.ra2_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:21:22Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-19T10:28:35Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir ecaresnet50t.ra2_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/ecaresnet50t.ra2_in1k-turbo-green-smashed --local-dir ecaresnet50t.ra2_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "ecaresnet50t.ra2_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "ecaresnet50t.ra2_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model ecaresnet50t.ra2_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/vit_base_patch32_384.augreg_in1k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:21:21Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-19T13:05:38Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir vit_base_patch32_384.augreg_in1k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/vit_base_patch32_384.augreg_in1k-turbo-tiny-green-smashed --local-dir vit_base_patch32_384.augreg_in1k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "vit_base_patch32_384.augreg_in1k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "vit_base_patch32_384.augreg_in1k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model vit_base_patch32_384.augreg_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/maxvit_large_tf_224.in21k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:21:19Z | 2 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T05:09:18Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir maxvit_large_tf_224.in21k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/maxvit_large_tf_224.in21k-turbo-tiny-green-smashed --local-dir maxvit_large_tf_224.in21k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "maxvit_large_tf_224.in21k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "maxvit_large_tf_224.in21k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model maxvit_large_tf_224.in21k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/wide_resnet101_2.tv2_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:21:13Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-14T11:42:05Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir wide_resnet101_2.tv2_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/wide_resnet101_2.tv2_in1k-turbo-green-smashed --local-dir wide_resnet101_2.tv2_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "wide_resnet101_2.tv2_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "wide_resnet101_2.tv2_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model wide_resnet101_2.tv2_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/vit_small_patch8_224.dino-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:21:12Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-14T11:31:16Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir vit_small_patch8_224.dino-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/vit_small_patch8_224.dino-turbo-tiny-green-smashed --local-dir vit_small_patch8_224.dino-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "vit_small_patch8_224.dino-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "vit_small_patch8_224.dino-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model vit_small_patch8_224.dino before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/flexivit_base.300ep_in21k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:21:11Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-19T11:30:26Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir flexivit_base.300ep_in21k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/flexivit_base.300ep_in21k-turbo-tiny-green-smashed --local-dir flexivit_base.300ep_in21k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "flexivit_base.300ep_in21k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "flexivit_base.300ep_in21k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model flexivit_base.300ep_in21k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/vit_huge_patch14_clip_224.laion2b_ft_in12k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:21:10Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-14T11:15:24Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir vit_huge_patch14_clip_224.laion2b_ft_in12k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/vit_huge_patch14_clip_224.laion2b_ft_in12k-turbo-tiny-green-smashed --local-dir vit_huge_patch14_clip_224.laion2b_ft_in12k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "vit_huge_patch14_clip_224.laion2b_ft_in12k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "vit_huge_patch14_clip_224.laion2b_ft_in12k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model vit_huge_patch14_clip_224.laion2b_ft_in12k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/resnext101_32x4d.fb_swsl_ig1b_ft_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:21:08Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-14T09:36:11Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir resnext101_32x4d.fb_swsl_ig1b_ft_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/resnext101_32x4d.fb_swsl_ig1b_ft_in1k-turbo-green-smashed --local-dir resnext101_32x4d.fb_swsl_ig1b_ft_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "resnext101_32x4d.fb_swsl_ig1b_ft_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "resnext101_32x4d.fb_swsl_ig1b_ft_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model resnext101_32x4d.fb_swsl_ig1b_ft_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/resnet14t.c3_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:21:06Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T08:45:02Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir resnet14t.c3_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/resnet14t.c3_in1k-turbo-green-smashed --local-dir resnet14t.c3_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "resnet14t.c3_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "resnet14t.c3_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model resnet14t.c3_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/resnet152d.gluon_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:21:05Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T09:26:33Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir resnet152d.gluon_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/resnet152d.gluon_in1k-turbo-green-smashed --local-dir resnet152d.gluon_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "resnet152d.gluon_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "resnet152d.gluon_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model resnet152d.gluon_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/convnext_tiny.in12k_ft_in1k_384-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:21:04Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-19T12:45:57Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir convnext_tiny.in12k_ft_in1k_384-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/convnext_tiny.in12k_ft_in1k_384-turbo-tiny-green-smashed --local-dir convnext_tiny.in12k_ft_in1k_384-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "convnext_tiny.in12k_ft_in1k_384-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "convnext_tiny.in12k_ft_in1k_384-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model convnext_tiny.in12k_ft_in1k_384 before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/resnet101.tv2_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:21:03Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T09:15:43Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir resnet101.tv2_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/resnet101.tv2_in1k-turbo-green-smashed --local-dir resnet101.tv2_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "resnet101.tv2_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "resnet101.tv2_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model resnet101.tv2_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/convnext_atto.d2_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:21:02Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-07T16:23:35Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir convnext_atto.d2_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/convnext_atto.d2_in1k-turbo-green-smashed --local-dir convnext_atto.d2_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "convnext_atto.d2_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "convnext_atto.d2_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model convnext_atto.d2_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/resnet152.gluon_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:21:01Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T09:23:37Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir resnet152.gluon_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/resnet152.gluon_in1k-turbo-green-smashed --local-dir resnet152.gluon_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "resnet152.gluon_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "resnet152.gluon_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model resnet152.gluon_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/seresnext101_32x8d.ah_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:20:57Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-14T09:58:06Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir seresnext101_32x8d.ah_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/seresnext101_32x8d.ah_in1k-turbo-green-smashed --local-dir seresnext101_32x8d.ah_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "seresnext101_32x8d.ah_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "seresnext101_32x8d.ah_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model seresnext101_32x8d.ah_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/resnetaa50d.sw_in12k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:20:55Z | 3 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T09:31:02Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir resnetaa50d.sw_in12k-turbo-green-smashed
huggingface-cli download PrunaAI/resnetaa50d.sw_in12k-turbo-green-smashed --local-dir resnetaa50d.sw_in12k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "resnetaa50d.sw_in12k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "resnetaa50d.sw_in12k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model resnetaa50d.sw_in12k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/convformer_s36.sail_in22k_ft_in1k_384-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:20:54Z | 3 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-19T12:42:13Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir convformer_s36.sail_in22k_ft_in1k_384-turbo-green-smashed
huggingface-cli download PrunaAI/convformer_s36.sail_in22k_ft_in1k_384-turbo-green-smashed --local-dir convformer_s36.sail_in22k_ft_in1k_384-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "convformer_s36.sail_in22k_ft_in1k_384-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "convformer_s36.sail_in22k_ft_in1k_384-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model convformer_s36.sail_in22k_ft_in1k_384 before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/vit_base_patch16_224.augreg_in21k_ft_in1k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:20:53Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-14T10:58:44Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir vit_base_patch16_224.augreg_in21k_ft_in1k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/vit_base_patch16_224.augreg_in21k_ft_in1k-turbo-tiny-green-smashed --local-dir vit_base_patch16_224.augreg_in21k_ft_in1k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "vit_base_patch16_224.augreg_in21k_ft_in1k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "vit_base_patch16_224.augreg_in21k_ft_in1k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model vit_base_patch16_224.augreg_in21k_ft_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/vit_tiny_r_s16_p8_224.augreg_in21k_ft_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:20:52Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-14T11:38:02Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir vit_tiny_r_s16_p8_224.augreg_in21k_ft_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/vit_tiny_r_s16_p8_224.augreg_in21k_ft_in1k-turbo-green-smashed --local-dir vit_tiny_r_s16_p8_224.augreg_in21k_ft_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "vit_tiny_r_s16_p8_224.augreg_in21k_ft_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "vit_tiny_r_s16_p8_224.augreg_in21k_ft_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model vit_tiny_r_s16_p8_224.augreg_in21k_ft_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/maxxvitv2_rmlp_base_rw_224.sw_in12k_ft_in1k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:20:51Z | 5 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T05:26:50Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir maxxvitv2_rmlp_base_rw_224.sw_in12k_ft_in1k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/maxxvitv2_rmlp_base_rw_224.sw_in12k_ft_in1k-turbo-tiny-green-smashed --local-dir maxxvitv2_rmlp_base_rw_224.sw_in12k_ft_in1k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "maxxvitv2_rmlp_base_rw_224.sw_in12k_ft_in1k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "maxxvitv2_rmlp_base_rw_224.sw_in12k_ft_in1k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model maxxvitv2_rmlp_base_rw_224.sw_in12k_ft_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/vit_large_patch14_clip_336.laion2b_ft_in1k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:20:45Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-19T12:29:34Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir vit_large_patch14_clip_336.laion2b_ft_in1k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/vit_large_patch14_clip_336.laion2b_ft_in1k-turbo-tiny-green-smashed --local-dir vit_large_patch14_clip_336.laion2b_ft_in1k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "vit_large_patch14_clip_336.laion2b_ft_in1k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "vit_large_patch14_clip_336.laion2b_ft_in1k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model vit_large_patch14_clip_336.laion2b_ft_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/gcvit_base.in1k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:20:44Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T03:17:56Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir gcvit_base.in1k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/gcvit_base.in1k-turbo-tiny-green-smashed --local-dir gcvit_base.in1k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "gcvit_base.in1k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "gcvit_base.in1k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model gcvit_base.in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/seresnext26ts.ch_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:20:41Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-14T09:55:05Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir seresnext26ts.ch_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/seresnext26ts.ch_in1k-turbo-green-smashed --local-dir seresnext26ts.ch_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "seresnext26ts.ch_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "seresnext26ts.ch_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model seresnext26ts.ch_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/vit_large_r50_s32_224.augreg_in21k_ft_in1k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:20:39Z | 2 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-14T11:28:39Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir vit_large_r50_s32_224.augreg_in21k_ft_in1k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/vit_large_r50_s32_224.augreg_in21k_ft_in1k-turbo-tiny-green-smashed --local-dir vit_large_r50_s32_224.augreg_in21k_ft_in1k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "vit_large_r50_s32_224.augreg_in21k_ft_in1k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "vit_large_r50_s32_224.augreg_in21k_ft_in1k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model vit_large_r50_s32_224.augreg_in21k_ft_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/vit_small_r26_s32_224.augreg_in21k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:20:37Z | 2 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-14T11:35:05Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir vit_small_r26_s32_224.augreg_in21k-turbo-green-smashed
huggingface-cli download PrunaAI/vit_small_r26_s32_224.augreg_in21k-turbo-green-smashed --local-dir vit_small_r26_s32_224.augreg_in21k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "vit_small_r26_s32_224.augreg_in21k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "vit_small_r26_s32_224.augreg_in21k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model vit_small_r26_s32_224.augreg_in21k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/vit_base_patch32_clip_384.openai_ft_in12k_in1k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:20:36Z | 2 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-19T13:07:58Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir vit_base_patch32_clip_384.openai_ft_in12k_in1k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/vit_base_patch32_clip_384.openai_ft_in12k_in1k-turbo-tiny-green-smashed --local-dir vit_base_patch32_clip_384.openai_ft_in12k_in1k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "vit_base_patch32_clip_384.openai_ft_in12k_in1k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "vit_base_patch32_clip_384.openai_ft_in12k_in1k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model vit_base_patch32_clip_384.openai_ft_in12k_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/flexivit_base.600ep_in1k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:20:36Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-19T11:31:52Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir flexivit_base.600ep_in1k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/flexivit_base.600ep_in1k-turbo-tiny-green-smashed --local-dir flexivit_base.600ep_in1k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "flexivit_base.600ep_in1k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "flexivit_base.600ep_in1k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model flexivit_base.600ep_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/flexivit_small.600ep_in1k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:20:35Z | 3 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-19T11:37:13Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir flexivit_small.600ep_in1k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/flexivit_small.600ep_in1k-turbo-tiny-green-smashed --local-dir flexivit_small.600ep_in1k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "flexivit_small.600ep_in1k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "flexivit_small.600ep_in1k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model flexivit_small.600ep_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/maxvit_tiny_tf_512.in1k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:20:33Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-19T13:35:36Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir maxvit_tiny_tf_512.in1k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/maxvit_tiny_tf_512.in1k-turbo-tiny-green-smashed --local-dir maxvit_tiny_tf_512.in1k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "maxvit_tiny_tf_512.in1k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "maxvit_tiny_tf_512.in1k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model maxvit_tiny_tf_512.in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/convnextv2_femto.fcmae-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:20:32Z | 2 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-07T17:42:30Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir convnextv2_femto.fcmae-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/convnextv2_femto.fcmae-turbo-tiny-green-smashed --local-dir convnextv2_femto.fcmae-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "convnextv2_femto.fcmae-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "convnextv2_femto.fcmae-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model convnextv2_femto.fcmae before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/resnet152.a3_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:20:31Z | 3 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T09:21:59Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir resnet152.a3_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/resnet152.a3_in1k-turbo-green-smashed --local-dir resnet152.a3_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "resnet152.a3_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "resnet152.a3_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model resnet152.a3_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/legacy_seresnet101.in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:20:30Z | 5 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T04:43:05Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir legacy_seresnet101.in1k-turbo-green-smashed
huggingface-cli download PrunaAI/legacy_seresnet101.in1k-turbo-green-smashed --local-dir legacy_seresnet101.in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "legacy_seresnet101.in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "legacy_seresnet101.in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model legacy_seresnet101.in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/resnetv2_50x1_bit.goog_in21k_ft_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:20:29Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T09:50:59Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir resnetv2_50x1_bit.goog_in21k_ft_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/resnetv2_50x1_bit.goog_in21k_ft_in1k-turbo-green-smashed --local-dir resnetv2_50x1_bit.goog_in21k_ft_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "resnetv2_50x1_bit.goog_in21k_ft_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "resnetv2_50x1_bit.goog_in21k_ft_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model resnetv2_50x1_bit.goog_in21k_ft_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/maxvit_xlarge_tf_512.in21k_ft_in1k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:20:27Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-19T13:44:28Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir maxvit_xlarge_tf_512.in21k_ft_in1k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/maxvit_xlarge_tf_512.in21k_ft_in1k-turbo-tiny-green-smashed --local-dir maxvit_xlarge_tf_512.in21k_ft_in1k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "maxvit_xlarge_tf_512.in21k_ft_in1k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "maxvit_xlarge_tf_512.in21k_ft_in1k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model maxvit_xlarge_tf_512.in21k_ft_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/resnet18.gluon_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:20:25Z | 2 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T08:47:33Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir resnet18.gluon_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/resnet18.gluon_in1k-turbo-green-smashed --local-dir resnet18.gluon_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "resnet18.gluon_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "resnet18.gluon_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model resnet18.gluon_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/vit_small_patch16_224.dino-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:20:24Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-14T11:32:38Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir vit_small_patch16_224.dino-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/vit_small_patch16_224.dino-turbo-tiny-green-smashed --local-dir vit_small_patch16_224.dino-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "vit_small_patch16_224.dino-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "vit_small_patch16_224.dino-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model vit_small_patch16_224.dino before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/convnext_large_mlp.clip_laion2b_augreg-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:20:23Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-07T16:55:15Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir convnext_large_mlp.clip_laion2b_augreg-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/convnext_large_mlp.clip_laion2b_augreg-turbo-tiny-green-smashed --local-dir convnext_large_mlp.clip_laion2b_augreg-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "convnext_large_mlp.clip_laion2b_augreg-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "convnext_large_mlp.clip_laion2b_augreg-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model convnext_large_mlp.clip_laion2b_augreg before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/resnet50d.a1_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:20:22Z | 2 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T09:06:59Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir resnet50d.a1_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/resnet50d.a1_in1k-turbo-green-smashed --local-dir resnet50d.a1_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "resnet50d.a1_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "resnet50d.a1_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model resnet50d.a1_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/convnext_nano.in12k_ft_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:20:20Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-07T17:15:32Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir convnext_nano.in12k_ft_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/convnext_nano.in12k_ft_in1k-turbo-green-smashed --local-dir convnext_nano.in12k_ft_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "convnext_nano.in12k_ft_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "convnext_nano.in12k_ft_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model convnext_nano.in12k_ft_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/flexivit_large.300ep_in1k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:20:19Z | 4 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-19T11:36:25Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir flexivit_large.300ep_in1k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/flexivit_large.300ep_in1k-turbo-tiny-green-smashed --local-dir flexivit_large.300ep_in1k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "flexivit_large.300ep_in1k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "flexivit_large.300ep_in1k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model flexivit_large.300ep_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/convnext_small.in12k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:20:18Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-07T17:26:06Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir convnext_small.in12k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/convnext_small.in12k-turbo-tiny-green-smashed --local-dir convnext_small.in12k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "convnext_small.in12k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "convnext_small.in12k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model convnext_small.in12k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/maxxvitv2_nano_rw_256.sw_in1k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:20:16Z | 3 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-19T11:47:27Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir maxxvitv2_nano_rw_256.sw_in1k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/maxxvitv2_nano_rw_256.sw_in1k-turbo-tiny-green-smashed --local-dir maxxvitv2_nano_rw_256.sw_in1k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "maxxvitv2_nano_rw_256.sw_in1k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "maxxvitv2_nano_rw_256.sw_in1k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model maxxvitv2_nano_rw_256.sw_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/vit_large_patch16_224.augreg_in21k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:20:09Z | 2 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-14T11:25:31Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir vit_large_patch16_224.augreg_in21k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/vit_large_patch16_224.augreg_in21k-turbo-tiny-green-smashed --local-dir vit_large_patch16_224.augreg_in21k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "vit_large_patch16_224.augreg_in21k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "vit_large_patch16_224.augreg_in21k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model vit_large_patch16_224.augreg_in21k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/resnext101_64x4d.gluon_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:20:07Z | 3 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-14T09:44:47Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir resnext101_64x4d.gluon_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/resnext101_64x4d.gluon_in1k-turbo-green-smashed --local-dir resnext101_64x4d.gluon_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "resnext101_64x4d.gluon_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "resnext101_64x4d.gluon_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model resnext101_64x4d.gluon_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/resnet50.a1h_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:20:06Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T08:56:55Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir resnet50.a1h_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/resnet50.a1h_in1k-turbo-green-smashed --local-dir resnet50.a1h_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "resnet50.a1h_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "resnet50.a1h_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model resnet50.a1h_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/resnet101-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:20:03Z | 2 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-07T13:37:32Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir resnet101-turbo-green-smashed
huggingface-cli download PrunaAI/resnet101-turbo-green-smashed --local-dir resnet101-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "resnet101-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "resnet101-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model resnet101 before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/vit_base_patch16_384.augreg_in21k_ft_in1k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:19:59Z | 2 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-19T13:05:12Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir vit_base_patch16_384.augreg_in21k_ft_in1k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/vit_base_patch16_384.augreg_in21k_ft_in1k-turbo-tiny-green-smashed --local-dir vit_base_patch16_384.augreg_in21k_ft_in1k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "vit_base_patch16_384.augreg_in21k_ft_in1k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "vit_base_patch16_384.augreg_in21k_ft_in1k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model vit_base_patch16_384.augreg_in21k_ft_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/convnext_base.fb_in22k_ft_in1k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:19:56Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-07T16:40:00Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir convnext_base.fb_in22k_ft_in1k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/convnext_base.fb_in22k_ft_in1k-turbo-tiny-green-smashed --local-dir convnext_base.fb_in22k_ft_in1k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "convnext_base.fb_in22k_ft_in1k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "convnext_base.fb_in22k_ft_in1k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model convnext_base.fb_in22k_ft_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/crossvit_18_dagger_408.in1k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:19:56Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-19T13:17:52Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir crossvit_18_dagger_408.in1k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/crossvit_18_dagger_408.in1k-turbo-tiny-green-smashed --local-dir crossvit_18_dagger_408.in1k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "crossvit_18_dagger_408.in1k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "crossvit_18_dagger_408.in1k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model crossvit_18_dagger_408.in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/resnet26d.bt_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:19:55Z | 2 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T08:50:09Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir resnet26d.bt_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/resnet26d.bt_in1k-turbo-green-smashed --local-dir resnet26d.bt_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "resnet26d.bt_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "resnet26d.bt_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model resnet26d.bt_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/tf_mobilenetv3_small_minimal_100.in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:19:54Z | 2 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-14T10:41:24Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir tf_mobilenetv3_small_minimal_100.in1k-turbo-green-smashed
huggingface-cli download PrunaAI/tf_mobilenetv3_small_minimal_100.in1k-turbo-green-smashed --local-dir tf_mobilenetv3_small_minimal_100.in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "tf_mobilenetv3_small_minimal_100.in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "tf_mobilenetv3_small_minimal_100.in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model tf_mobilenetv3_small_minimal_100.in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/resnet34.a3_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:19:52Z | 2 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T08:53:45Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir resnet34.a3_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/resnet34.a3_in1k-turbo-green-smashed --local-dir resnet34.a3_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "resnet34.a3_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "resnet34.a3_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model resnet34.a3_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/convnextv2_femto.fcmae_ft_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:19:51Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-07T17:42:45Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir convnextv2_femto.fcmae_ft_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/convnextv2_femto.fcmae_ft_in1k-turbo-green-smashed --local-dir convnextv2_femto.fcmae_ft_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "convnextv2_femto.fcmae_ft_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "convnextv2_femto.fcmae_ft_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model convnextv2_femto.fcmae_ft_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/convnext_pico.d1_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:19:49Z | 3 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-07T17:17:30Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir convnext_pico.d1_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/convnext_pico.d1_in1k-turbo-green-smashed --local-dir convnext_pico.d1_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "convnext_pico.d1_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "convnext_pico.d1_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model convnext_pico.d1_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/mobilenet_v3_small-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:19:48Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-07T13:26:53Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir mobilenet_v3_small-turbo-green-smashed
huggingface-cli download PrunaAI/mobilenet_v3_small-turbo-green-smashed --local-dir mobilenet_v3_small-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "mobilenet_v3_small-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "mobilenet_v3_small-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model mobilenet_v3_small before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/ecaresnet26t.ra2_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:19:47Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-19T10:20:43Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir ecaresnet26t.ra2_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/ecaresnet26t.ra2_in1k-turbo-green-smashed --local-dir ecaresnet26t.ra2_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "ecaresnet26t.ra2_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "ecaresnet26t.ra2_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model ecaresnet26t.ra2_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/maxvit_tiny_tf_384.in1k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:19:46Z | 2 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-19T12:55:44Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir maxvit_tiny_tf_384.in1k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/maxvit_tiny_tf_384.in1k-turbo-tiny-green-smashed --local-dir maxvit_tiny_tf_384.in1k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "maxvit_tiny_tf_384.in1k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "maxvit_tiny_tf_384.in1k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model maxvit_tiny_tf_384.in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/mobilenetv3_large_100.miil_in21k_ft_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:19:45Z | 3 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T05:45:19Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir mobilenetv3_large_100.miil_in21k_ft_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/mobilenetv3_large_100.miil_in21k_ft_in1k-turbo-green-smashed --local-dir mobilenetv3_large_100.miil_in21k_ft_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "mobilenetv3_large_100.miil_in21k_ft_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "mobilenetv3_large_100.miil_in21k_ft_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model mobilenetv3_large_100.miil_in21k_ft_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/vit_large_patch32_224.orig_in21k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:19:44Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-14T11:27:03Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir vit_large_patch32_224.orig_in21k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/vit_large_patch32_224.orig_in21k-turbo-tiny-green-smashed --local-dir vit_large_patch32_224.orig_in21k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "vit_large_patch32_224.orig_in21k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "vit_large_patch32_224.orig_in21k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model vit_large_patch32_224.orig_in21k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/vit_l_32-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:19:43Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-07T13:56:35Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir vit_l_32-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/vit_l_32-turbo-tiny-green-smashed --local-dir vit_l_32-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "vit_l_32-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "vit_l_32-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model vit_l_32 before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/mvitv2_large_cls.fb_inw21k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:19:42Z | 3 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T06:24:01Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir mvitv2_large_cls.fb_inw21k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/mvitv2_large_cls.fb_inw21k-turbo-tiny-green-smashed --local-dir mvitv2_large_cls.fb_inw21k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "mvitv2_large_cls.fb_inw21k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "mvitv2_large_cls.fb_inw21k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model mvitv2_large_cls.fb_inw21k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/convnext_atto_ols.a2_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:19:41Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-07T16:23:27Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir convnext_atto_ols.a2_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/convnext_atto_ols.a2_in1k-turbo-green-smashed --local-dir convnext_atto_ols.a2_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "convnext_atto_ols.a2_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "convnext_atto_ols.a2_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model convnext_atto_ols.a2_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/resnet101s.gluon_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:19:38Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T09:19:18Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir resnet101s.gluon_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/resnet101s.gluon_in1k-turbo-green-smashed --local-dir resnet101s.gluon_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "resnet101s.gluon_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "resnet101s.gluon_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model resnet101s.gluon_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/mvitv2_huge_cls.fb_inw21k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:19:36Z | 3 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T06:28:34Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir mvitv2_huge_cls.fb_inw21k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/mvitv2_huge_cls.fb_inw21k-turbo-tiny-green-smashed --local-dir mvitv2_huge_cls.fb_inw21k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "mvitv2_huge_cls.fb_inw21k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "mvitv2_huge_cls.fb_inw21k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model mvitv2_huge_cls.fb_inw21k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/convformer_s18.sail_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:19:34Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-19T10:03:11Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir convformer_s18.sail_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/convformer_s18.sail_in1k-turbo-green-smashed --local-dir convformer_s18.sail_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "convformer_s18.sail_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "convformer_s18.sail_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model convformer_s18.sail_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/ecaresnet50d_pruned.miil_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:19:33Z | 2 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-19T10:23:27Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir ecaresnet50d_pruned.miil_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/ecaresnet50d_pruned.miil_in1k-turbo-green-smashed --local-dir ecaresnet50d_pruned.miil_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "ecaresnet50d_pruned.miil_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "ecaresnet50d_pruned.miil_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model ecaresnet50d_pruned.miil_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/convformer_s36.sail_in1k_384-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:19:32Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-19T12:42:45Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir convformer_s36.sail_in1k_384-turbo-green-smashed
huggingface-cli download PrunaAI/convformer_s36.sail_in1k_384-turbo-green-smashed --local-dir convformer_s36.sail_in1k_384-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "convformer_s36.sail_in1k_384-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "convformer_s36.sail_in1k_384-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model convformer_s36.sail_in1k_384 before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/resnet50.tv_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:19:30Z | 2 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-14T09:15:06Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir resnet50.tv_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/resnet50.tv_in1k-turbo-green-smashed --local-dir resnet50.tv_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "resnet50.tv_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "resnet50.tv_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model resnet50.tv_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/resnetblur50.bt_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:19:29Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T09:34:19Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir resnetblur50.bt_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/resnetblur50.bt_in1k-turbo-green-smashed --local-dir resnetblur50.bt_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "resnetblur50.bt_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "resnetblur50.bt_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model resnetblur50.bt_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/vit_large_patch14_clip_224.laion2b_ft_in1k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:19:26Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-14T11:19:57Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir vit_large_patch14_clip_224.laion2b_ft_in1k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/vit_large_patch14_clip_224.laion2b_ft_in1k-turbo-tiny-green-smashed --local-dir vit_large_patch14_clip_224.laion2b_ft_in1k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "vit_large_patch14_clip_224.laion2b_ft_in1k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "vit_large_patch14_clip_224.laion2b_ft_in1k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model vit_large_patch14_clip_224.laion2b_ft_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/convnextv2_large.fcmae_ft_in1k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:19:23Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-07T17:48:18Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir convnextv2_large.fcmae_ft_in1k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/convnextv2_large.fcmae_ft_in1k-turbo-tiny-green-smashed --local-dir convnextv2_large.fcmae_ft_in1k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "convnextv2_large.fcmae_ft_in1k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "convnextv2_large.fcmae_ft_in1k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model convnextv2_large.fcmae_ft_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/convnext_large_mlp.clip_laion2b_soup_ft_in12k_384-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:19:22Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-19T12:46:31Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir convnext_large_mlp.clip_laion2b_soup_ft_in12k_384-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/convnext_large_mlp.clip_laion2b_soup_ft_in12k_384-turbo-tiny-green-smashed --local-dir convnext_large_mlp.clip_laion2b_soup_ft_in12k_384-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "convnext_large_mlp.clip_laion2b_soup_ft_in12k_384-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "convnext_large_mlp.clip_laion2b_soup_ft_in12k_384-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model convnext_large_mlp.clip_laion2b_soup_ft_in12k_384 before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/resnet50.ra_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:19:16Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T09:03:54Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir resnet50.ra_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/resnet50.ra_in1k-turbo-green-smashed --local-dir resnet50.ra_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "resnet50.ra_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "resnet50.ra_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model resnet50.ra_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/convnext_small.in12k_ft_in1k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:19:15Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-07T17:25:54Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir convnext_small.in12k_ft_in1k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/convnext_small.in12k_ft_in1k-turbo-tiny-green-smashed --local-dir convnext_small.in12k_ft_in1k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "convnext_small.in12k_ft_in1k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "convnext_small.in12k_ft_in1k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model convnext_small.in12k_ft_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/seresnextaa101d_32x8d.sw_in12k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:19:13Z | 2 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-14T10:00:51Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir seresnextaa101d_32x8d.sw_in12k-turbo-green-smashed
huggingface-cli download PrunaAI/seresnextaa101d_32x8d.sw_in12k-turbo-green-smashed --local-dir seresnextaa101d_32x8d.sw_in12k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "seresnextaa101d_32x8d.sw_in12k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "seresnextaa101d_32x8d.sw_in12k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model seresnextaa101d_32x8d.sw_in12k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/vit_medium_patch16_gap_256.sw_in12k_ft_in1k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:19:09Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-19T11:59:44Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir vit_medium_patch16_gap_256.sw_in12k_ft_in1k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/vit_medium_patch16_gap_256.sw_in12k_ft_in1k-turbo-tiny-green-smashed --local-dir vit_medium_patch16_gap_256.sw_in12k_ft_in1k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "vit_medium_patch16_gap_256.sw_in12k_ft_in1k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "vit_medium_patch16_gap_256.sw_in12k_ft_in1k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model vit_medium_patch16_gap_256.sw_in12k_ft_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/vit_medium_patch16_gap_240.sw_in12k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:19:08Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-19T11:28:20Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir vit_medium_patch16_gap_240.sw_in12k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/vit_medium_patch16_gap_240.sw_in12k-turbo-tiny-green-smashed --local-dir vit_medium_patch16_gap_240.sw_in12k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "vit_medium_patch16_gap_240.sw_in12k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "vit_medium_patch16_gap_240.sw_in12k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model vit_medium_patch16_gap_240.sw_in12k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/tresnet_l.miil_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:19:07Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-14T10:45:44Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir tresnet_l.miil_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/tresnet_l.miil_in1k-turbo-green-smashed --local-dir tresnet_l.miil_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "tresnet_l.miil_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "tresnet_l.miil_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model tresnet_l.miil_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/resnetrs420.tf_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:19:07Z | 2 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T09:52:48Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir resnetrs420.tf_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/resnetrs420.tf_in1k-turbo-green-smashed --local-dir resnetrs420.tf_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "resnetrs420.tf_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "resnetrs420.tf_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model resnetrs420.tf_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/resnet101d.gluon_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:19:06Z | 2 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T09:17:11Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir resnet101d.gluon_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/resnet101d.gluon_in1k-turbo-green-smashed --local-dir resnet101d.gluon_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "resnet101d.gluon_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "resnet101d.gluon_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model resnet101d.gluon_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/maxvit_base_tf_512.in21k_ft_in1k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:19:05Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-19T13:36:55Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir maxvit_base_tf_512.in21k_ft_in1k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/maxvit_base_tf_512.in21k_ft_in1k-turbo-tiny-green-smashed --local-dir maxvit_base_tf_512.in21k_ft_in1k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "maxvit_base_tf_512.in21k_ft_in1k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "maxvit_base_tf_512.in21k_ft_in1k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model maxvit_base_tf_512.in21k_ft_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/resnet50.bt_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:19:04Z | 4 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T08:59:53Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir resnet50.bt_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/resnet50.bt_in1k-turbo-green-smashed --local-dir resnet50.bt_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "resnet50.bt_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "resnet50.bt_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model resnet50.bt_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/vit_base_patch32_clip_224.openai_ft_in1k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:19:02Z | 4 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-14T11:08:47Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir vit_base_patch32_clip_224.openai_ft_in1k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/vit_base_patch32_clip_224.openai_ft_in1k-turbo-tiny-green-smashed --local-dir vit_base_patch32_clip_224.openai_ft_in1k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "vit_base_patch32_clip_224.openai_ft_in1k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "vit_base_patch32_clip_224.openai_ft_in1k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model vit_base_patch32_clip_224.openai_ft_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/resnet101.a2_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:19:01Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T09:13:44Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir resnet101.a2_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/resnet101.a2_in1k-turbo-green-smashed --local-dir resnet101.a2_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "resnet101.a2_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "resnet101.a2_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model resnet101.a2_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/resnetaa50d.d_in12k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:18:59Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T09:29:26Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir resnetaa50d.d_in12k-turbo-green-smashed
huggingface-cli download PrunaAI/resnetaa50d.d_in12k-turbo-green-smashed --local-dir resnetaa50d.d_in12k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "resnetaa50d.d_in12k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "resnetaa50d.d_in12k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model resnetaa50d.d_in12k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/mobilenetv3_rw.rmsp_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:18:56Z | 3 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T05:46:18Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir mobilenetv3_rw.rmsp_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/mobilenetv3_rw.rmsp_in1k-turbo-green-smashed --local-dir mobilenetv3_rw.rmsp_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "mobilenetv3_rw.rmsp_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "mobilenetv3_rw.rmsp_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model mobilenetv3_rw.rmsp_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/convnext_large_mlp.clip_laion2b_augreg_ft_in1k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:18:56Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-07T16:59:41Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir convnext_large_mlp.clip_laion2b_augreg_ft_in1k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/convnext_large_mlp.clip_laion2b_augreg_ft_in1k-turbo-tiny-green-smashed --local-dir convnext_large_mlp.clip_laion2b_augreg_ft_in1k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "convnext_large_mlp.clip_laion2b_augreg_ft_in1k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "convnext_large_mlp.clip_laion2b_augreg_ft_in1k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model convnext_large_mlp.clip_laion2b_augreg_ft_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/vit_large_patch14_clip_224.openai-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:18:54Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-14T11:22:45Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir vit_large_patch14_clip_224.openai-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/vit_large_patch14_clip_224.openai-turbo-tiny-green-smashed --local-dir vit_large_patch14_clip_224.openai-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "vit_large_patch14_clip_224.openai-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "vit_large_patch14_clip_224.openai-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model vit_large_patch14_clip_224.openai before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/resnetaa101d.sw_in12k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:18:53Z | 2 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T09:32:40Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir resnetaa101d.sw_in12k-turbo-green-smashed
huggingface-cli download PrunaAI/resnetaa101d.sw_in12k-turbo-green-smashed --local-dir resnetaa101d.sw_in12k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "resnetaa101d.sw_in12k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "resnetaa101d.sw_in12k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model resnetaa101d.sw_in12k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/vit_base_patch16_clip_384.laion2b_ft_in12k_in1k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:18:50Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-19T13:05:40Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir vit_base_patch16_clip_384.laion2b_ft_in12k_in1k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/vit_base_patch16_clip_384.laion2b_ft_in12k_in1k-turbo-tiny-green-smashed --local-dir vit_base_patch16_clip_384.laion2b_ft_in12k_in1k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "vit_base_patch16_clip_384.laion2b_ft_in12k_in1k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "vit_base_patch16_clip_384.laion2b_ft_in12k_in1k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model vit_base_patch16_clip_384.laion2b_ft_in12k_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/vit_large_patch16_384.augreg_in21k_ft_in1k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:18:49Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-19T13:09:49Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir vit_large_patch16_384.augreg_in21k_ft_in1k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/vit_large_patch16_384.augreg_in21k_ft_in1k-turbo-tiny-green-smashed --local-dir vit_large_patch16_384.augreg_in21k_ft_in1k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "vit_large_patch16_384.augreg_in21k_ft_in1k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "vit_large_patch16_384.augreg_in21k_ft_in1k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model vit_large_patch16_384.augreg_in21k_ft_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/vit_base_patch8_224.dino-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:18:47Z | 3 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-14T10:57:38Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir vit_base_patch8_224.dino-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/vit_base_patch8_224.dino-turbo-tiny-green-smashed --local-dir vit_base_patch8_224.dino-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "vit_base_patch8_224.dino-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "vit_base_patch8_224.dino-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model vit_base_patch8_224.dino before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/vit_base_patch16_224.augreg_in21k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:18:46Z | 2 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-14T10:58:13Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir vit_base_patch16_224.augreg_in21k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/vit_base_patch16_224.augreg_in21k-turbo-tiny-green-smashed --local-dir vit_base_patch16_224.augreg_in21k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "vit_base_patch16_224.augreg_in21k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "vit_base_patch16_224.augreg_in21k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model vit_base_patch16_224.augreg_in21k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/convnext_large_mlp.clip_laion2b_augreg_ft_in12k_384-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:18:45Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-19T12:45:21Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir convnext_large_mlp.clip_laion2b_augreg_ft_in12k_384-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/convnext_large_mlp.clip_laion2b_augreg_ft_in12k_384-turbo-tiny-green-smashed --local-dir convnext_large_mlp.clip_laion2b_augreg_ft_in12k_384-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "convnext_large_mlp.clip_laion2b_augreg_ft_in12k_384-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "convnext_large_mlp.clip_laion2b_augreg_ft_in12k_384-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model convnext_large_mlp.clip_laion2b_augreg_ft_in12k_384 before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/convnext_tiny.in12k-turbo-tiny-green-smashed
|
PrunaAI
| 2024-11-13T13:18:43Z | 1 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-07T17:28:54Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir convnext_tiny.in12k-turbo-tiny-green-smashed
huggingface-cli download PrunaAI/convnext_tiny.in12k-turbo-tiny-green-smashed --local-dir convnext_tiny.in12k-turbo-tiny-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "convnext_tiny.in12k-turbo-tiny-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "convnext_tiny.in12k-turbo-tiny-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model convnext_tiny.in12k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/resnet101.a1h_in1k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:18:42Z | 2 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-10T09:12:13Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir resnet101.a1h_in1k-turbo-green-smashed
huggingface-cli download PrunaAI/resnet101.a1h_in1k-turbo-green-smashed --local-dir resnet101.a1h_in1k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "resnet101.a1h_in1k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "resnet101.a1h_in1k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model resnet101.a1h_in1k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
PrunaAI/convnext_nano.in12k-turbo-green-smashed
|
PrunaAI
| 2024-11-13T13:18:41Z | 2 | 0 |
pruna-engine
|
[
"pruna-engine",
"region:us"
] | null | 2024-03-07T17:14:20Z |
---
library_name: pruna-engine
thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
metrics:
- memory_disk
- memory_inference
- inference_latency
- inference_throughput
- inference_CO2_emissions
- inference_energy_consumption
---
<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<a href="https://docs.pruna.ai/en/latest/setup/pip.html" target="_blank" rel="noopener noreferrer">
<img src="https://imgur.com/rVAgqMY.png" alt="PrunaAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</a>
</div>
<!-- header end -->
[](https://twitter.com/PrunaAI)
[](https://github.com/PrunaAI)
[](https://www.linkedin.com/company/93832878/admin/feed/posts/?feedType=following)
[](https://discord.gg/rskEr4BZJx)
<div style="color: #9B1DBE; font-size: 2em; font-weight: bold;">
Deprecation Notice: This model is deprecated and will no longer receive updates.
</div>
<br><br>
# Simply make AI models cheaper, smaller, faster, and greener!
- Give a thumbs up if you like this model!
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your *own* AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- Read the documentations to know more [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/)
- Join Pruna AI community on Discord [here](https://discord.gg/rskEr4BZJx) to share feedback/suggestions or get help.
## Results
**Frequently Asked Questions**
- ***How does the compression work?*** The model is compressed by combining quantization, xformers, jit, cuda graphs, triton.
- ***How does the model quality change?*** The quality of the model output might slightly vary compared to the base model.
- ***How is the model efficiency evaluated?*** These results were obtained on NVIDIA A100-PCIE-40GB with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
- ***What is the model format?*** We used a custom Pruna model format based on pickle to make models compatible with the compression methods. We provide a tutorial to run models in dockers in the documentation [here](https://pruna-ai-pruna.readthedocs-hosted.com/en/latest/) if needed.
- ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
- ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
- ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
- ***What are "Sync" and "Async" metrics?*** "Sync" metrics are obtained by syncing all GPU processes and stop measurement when all of them are executed. "Async" metrics are obtained without syncing all GPU processes and stop when the model output can be used by the CPU. We provide both metrics since both could be relevant depending on the use-case. We recommend to test the efficiency gains directly in your use-cases.
## Setup
You can run the smashed model with these steps:
0. Check that you have linux, python 3.10, and cuda 12.1.0 requirements installed. For cuda, check with `nvcc --version` and install with `conda install nvidia/label/cuda-12.1.0::cuda`.
1. Install the `pruna-engine` available [here](https://pypi.org/project/pruna-engine/) on Pypi. It might take up to 15 minutes to install.
```bash
pip install pruna-engine[gpu]==0.7.1 --extra-index-url https://pypi.nvidia.com --extra-index-url https://pypi.ngc.nvidia.com --extra-index-url https://prunaai.pythonanywhere.com/
```
2. Download the model files using one of these three options.
- Option 1 - Use command line interface (CLI):
```bash
mkdir convnext_nano.in12k-turbo-green-smashed
huggingface-cli download PrunaAI/convnext_nano.in12k-turbo-green-smashed --local-dir convnext_nano.in12k-turbo-green-smashed --local-dir-use-symlinks False
```
- Option 2 - Use Python:
```python
import subprocess
repo_name = "convnext_nano.in12k-turbo-green-smashed"
subprocess.run(["mkdir", repo_name])
subprocess.run(["huggingface-cli", "download", 'PrunaAI/'+ repo_name, "--local-dir", repo_name, "--local-dir-use-symlinks", "False"])
```
- Option 3 - Download them manually on the HuggingFace model page.
3. Load & run the model.
```python
from pruna_engine.PrunaModel import PrunaModel
model_path = "convnext_nano.in12k-turbo-green-smashed/model" # Specify the downloaded model path.
smashed_model = PrunaModel.load_model(model_path) # Load the model.
import torch; image = torch.rand(1, 3, 224, 224).to('cuda')
smashed_model(image)
```
## Configurations
The configuration info are in `model/smash_config.json`.
## Credits & License
The license of the smashed model follows the license of the original model. Please check the license of the original model convnext_nano.in12k before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
## Want to compress other models?
- Contact us and tell us which model to compress next [here](https://www.pruna.ai/contact).
- Request access to easily compress your own AI models [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
|
Subsets and Splits
Filtered Qwen2.5 Distill Models
Identifies specific configurations of models by filtering cards that contain 'distill', 'qwen2.5', '7b' while excluding certain base models and incorrect model ID patterns, uncovering unique model variants.
Filtered Model Cards Count
Finds the count of entries with specific card details that include 'distill', 'qwen2.5', '7b' but exclude certain base models, revealing valuable insights about the dataset's content distribution.
Filtered Distill Qwen 7B Models
Filters for specific card entries containing 'distill', 'qwen', and '7b', excluding certain strings and patterns, to identify relevant model configurations.
Filtered Qwen-7b Model Cards
The query performs a detailed filtering based on specific keywords and excludes certain entries, which could be useful for identifying a specific subset of cards but does not provide deeper insights or trends.
Filtered Qwen 7B Model Cards
The query filters for specific terms related to "distilled" or "distill", "qwen", and "7b" in the 'card' column but excludes certain base models, providing a limited set of entries for further inspection.
Qwen 7B Distilled Models
The query provides a basic filtering of records to find specific card names that include keywords related to distilled Qwen 7b models, excluding a particular base model, which gives limited insight but helps in focusing on relevant entries.
Qwen 7B Distilled Model Cards
The query filters data based on specific keywords in the modelId and card fields, providing limited insight primarily useful for locating specific entries rather than revealing broad patterns or trends.
Qwen 7B Distilled Models
Finds all entries containing the terms 'distilled', 'qwen', and '7b' in a case-insensitive manner, providing a filtered set of records but without deeper analysis.
Distilled Qwen 7B Models
The query filters for specific model IDs containing 'distilled', 'qwen', and '7b', providing a basic retrieval of relevant entries but without deeper analysis or insight.
Filtered Model Cards with Distill Qwen2.
Filters and retrieves records containing specific keywords in the card description while excluding certain phrases, providing a basic count of relevant entries.
Filtered Model Cards with Distill Qwen 7
The query filters specific variations of card descriptions containing 'distill', 'qwen', and '7b' while excluding a particular base model, providing limited but specific data retrieval.
Distill Qwen 7B Model Cards
The query filters and retrieves rows where the 'card' column contains specific keywords ('distill', 'qwen', and '7b'), providing a basic filter result that can help in identifying specific entries.