Hugging Face's logo

qualcomm
/
AOT-GAN

Image-to-Image
PyTorch
LiteRT
ONNX
android
Model card Files Community
1
bhushans commited on
Commit
1c8268d
·
verified ·
1 Parent(s): bb3eed3

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +40 -22
README.md CHANGED
@@ -34,25 +34,27 @@ More details on model performance across various devices, can be found
34
 
35
  | Model | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model
36
  |---|---|---|---|---|---|---|---|---|
37
- | AOT-GAN | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 152.763 ms | 3 - 6 MB | FP16 | NPU | [AOT-GAN.tflite](https://huggingface.co/qualcomm/AOT-GAN/blob/main/AOT-GAN.tflite) |
38
- | AOT-GAN | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | QNN | 153.27 ms | 4 - 22 MB | FP16 | NPU | [AOT-GAN.so](https://huggingface.co/qualcomm/AOT-GAN/blob/main/AOT-GAN.so) |
39
- | AOT-GAN | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 112.21 ms | 0 - 215 MB | FP16 | NPU | [AOT-GAN.tflite](https://huggingface.co/qualcomm/AOT-GAN/blob/main/AOT-GAN.tflite) |
40
- | AOT-GAN | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | QNN | 112.615 ms | 4 - 65 MB | FP16 | NPU | [AOT-GAN.so](https://huggingface.co/qualcomm/AOT-GAN/blob/main/AOT-GAN.so) |
41
- | AOT-GAN | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 118.64 ms | 0 - 83 MB | FP16 | NPU | [AOT-GAN.tflite](https://huggingface.co/qualcomm/AOT-GAN/blob/main/AOT-GAN.tflite) |
42
- | AOT-GAN | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | QNN | 118.61 ms | 3 - 65 MB | FP16 | NPU | Use Export Script |
43
- | AOT-GAN | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 149.861 ms | 3 - 5 MB | FP16 | NPU | [AOT-GAN.tflite](https://huggingface.co/qualcomm/AOT-GAN/blob/main/AOT-GAN.tflite) |
44
- | AOT-GAN | QCS8550 (Proxy) | QCS8550 Proxy | QNN | 92.085 ms | 4 - 5 MB | FP16 | NPU | Use Export Script |
45
- | AOT-GAN | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 153.081 ms | 3 - 5 MB | FP16 | NPU | [AOT-GAN.tflite](https://huggingface.co/qualcomm/AOT-GAN/blob/main/AOT-GAN.tflite) |
46
- | AOT-GAN | SA8255 (Proxy) | SA8255P Proxy | QNN | 92.57 ms | 4 - 5 MB | FP16 | NPU | Use Export Script |
47
- | AOT-GAN | SA8775 (Proxy) | SA8775P Proxy | TFLITE | 152.952 ms | 3 - 5 MB | FP16 | NPU | [AOT-GAN.tflite](https://huggingface.co/qualcomm/AOT-GAN/blob/main/AOT-GAN.tflite) |
48
- | AOT-GAN | SA8775 (Proxy) | SA8775P Proxy | QNN | 92.932 ms | 4 - 5 MB | FP16 | NPU | Use Export Script |
49
- | AOT-GAN | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 152.762 ms | 4 - 6 MB | FP16 | NPU | [AOT-GAN.tflite](https://huggingface.co/qualcomm/AOT-GAN/blob/main/AOT-GAN.tflite) |
50
- | AOT-GAN | SA8650 (Proxy) | SA8650P Proxy | QNN | 92.47 ms | 5 - 6 MB | FP16 | NPU | Use Export Script |
51
- | AOT-GAN | SA8295P ADP | SA8295P | TFLITE | 218.814 ms | 3 - 66 MB | FP16 | NPU | [AOT-GAN.tflite](https://huggingface.co/qualcomm/AOT-GAN/blob/main/AOT-GAN.tflite) |
52
- | AOT-GAN | SA8295P ADP | SA8295P | QNN | 164.324 ms | 1 - 7 MB | FP16 | NPU | Use Export Script |
53
- | AOT-GAN | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 194.495 ms | 1 - 185 MB | FP16 | NPU | [AOT-GAN.tflite](https://huggingface.co/qualcomm/AOT-GAN/blob/main/AOT-GAN.tflite) |
54
- | AOT-GAN | QCS8450 (Proxy) | QCS8450 Proxy | QNN | 196.67 ms | 3 - 49 MB | FP16 | NPU | Use Export Script |
55
- | AOT-GAN | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 96.297 ms | 4 - 4 MB | FP16 | NPU | Use Export Script |
 
 
56
 
57
 
58
 
@@ -116,8 +118,8 @@ Profiling Results
116
  AOT-GAN
117
  Device : Samsung Galaxy S23 (13)
118
  Runtime : TFLITE
119
- Estimated inference time (ms) : 152.8
120
- Estimated peak memory usage (MB): [3, 6]
121
  Total # Ops : 235
122
  Compute Unit(s) : NPU (235 ops)
123
  ```
@@ -138,13 +140,29 @@ in memory using the `jit.trace` and then call the `submit_compile_job` API.
138
  import torch
139
 
140
  import qai_hub as hub
141
- from qai_hub_models.models.aotgan import
142
 
143
  # Load the model
 
144
 
145
  # Device
146
  device = hub.Device("Samsung Galaxy S23")
147
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
148
 
149
  ```
150
 
 
34
 
35
  | Model | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model
36
  |---|---|---|---|---|---|---|---|---|
37
+ | AOT-GAN | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 153.257 ms | 3 - 36 MB | FP16 | NPU | [AOT-GAN.tflite](https://huggingface.co/qualcomm/AOT-GAN/blob/main/AOT-GAN.tflite) |
38
+ | AOT-GAN | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | QNN | 153.759 ms | 0 - 36 MB | FP16 | NPU | [AOT-GAN.so](https://huggingface.co/qualcomm/AOT-GAN/blob/main/AOT-GAN.so) |
39
+ | AOT-GAN | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 112.139 ms | 1 - 61 MB | FP16 | NPU | [AOT-GAN.tflite](https://huggingface.co/qualcomm/AOT-GAN/blob/main/AOT-GAN.tflite) |
40
+ | AOT-GAN | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | QNN | 111.92 ms | 3 - 61 MB | FP16 | NPU | [AOT-GAN.so](https://huggingface.co/qualcomm/AOT-GAN/blob/main/AOT-GAN.so) |
41
+ | AOT-GAN | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 118.38 ms | 3 - 65 MB | FP16 | NPU | [AOT-GAN.tflite](https://huggingface.co/qualcomm/AOT-GAN/blob/main/AOT-GAN.tflite) |
42
+ | AOT-GAN | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | QNN | 118.426 ms | 4 - 66 MB | FP16 | NPU | Use Export Script |
43
+ | AOT-GAN | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 153.157 ms | 3 - 30 MB | FP16 | NPU | [AOT-GAN.tflite](https://huggingface.co/qualcomm/AOT-GAN/blob/main/AOT-GAN.tflite) |
44
+ | AOT-GAN | QCS8550 (Proxy) | QCS8550 Proxy | QNN | 102.102 ms | 4 - 5 MB | FP16 | NPU | Use Export Script |
45
+ | AOT-GAN | SA7255P ADP | SA7255P | TFLITE | 3625.285 ms | 3 - 64 MB | FP16 | NPU | [AOT-GAN.tflite](https://huggingface.co/qualcomm/AOT-GAN/blob/main/AOT-GAN.tflite) |
46
+ | AOT-GAN | SA7255P ADP | SA7255P | QNN | 3580.513 ms | 3 - 10 MB | FP16 | NPU | Use Export Script |
47
+ | AOT-GAN | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 153.372 ms | 3 - 31 MB | FP16 | NPU | [AOT-GAN.tflite](https://huggingface.co/qualcomm/AOT-GAN/blob/main/AOT-GAN.tflite) |
48
+ | AOT-GAN | SA8255 (Proxy) | SA8255P Proxy | QNN | 102.12 ms | 4 - 5 MB | FP16 | NPU | Use Export Script |
49
+ | AOT-GAN | SA8295P ADP | SA8295P | TFLITE | 219.091 ms | 3 - 51 MB | FP16 | NPU | [AOT-GAN.tflite](https://huggingface.co/qualcomm/AOT-GAN/blob/main/AOT-GAN.tflite) |
50
+ | AOT-GAN | SA8295P ADP | SA8295P | QNN | 164.549 ms | 1 - 7 MB | FP16 | NPU | Use Export Script |
51
+ | AOT-GAN | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 153.38 ms | 3 - 34 MB | FP16 | NPU | [AOT-GAN.tflite](https://huggingface.co/qualcomm/AOT-GAN/blob/main/AOT-GAN.tflite) |
52
+ | AOT-GAN | SA8650 (Proxy) | SA8650P Proxy | QNN | 101.825 ms | 3 - 6 MB | FP16 | NPU | Use Export Script |
53
+ | AOT-GAN | SA8775P ADP | SA8775P | TFLITE | 241.89 ms | 3 - 64 MB | FP16 | NPU | [AOT-GAN.tflite](https://huggingface.co/qualcomm/AOT-GAN/blob/main/AOT-GAN.tflite) |
54
+ | AOT-GAN | SA8775P ADP | SA8775P | QNN | 181.587 ms | 3 - 9 MB | FP16 | NPU | Use Export Script |
55
+ | AOT-GAN | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 196.585 ms | 3 - 53 MB | FP16 | NPU | [AOT-GAN.tflite](https://huggingface.co/qualcomm/AOT-GAN/blob/main/AOT-GAN.tflite) |
56
+ | AOT-GAN | QCS8450 (Proxy) | QCS8450 Proxy | QNN | 196.144 ms | 4 - 53 MB | FP16 | NPU | Use Export Script |
57
+ | AOT-GAN | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 110.93 ms | 4 - 4 MB | FP16 | NPU | Use Export Script |
58
 
59
 
60
 
 
118
  AOT-GAN
119
  Device : Samsung Galaxy S23 (13)
120
  Runtime : TFLITE
121
+ Estimated inference time (ms) : 153.3
122
+ Estimated peak memory usage (MB): [3, 36]
123
  Total # Ops : 235
124
  Compute Unit(s) : NPU (235 ops)
125
  ```
 
140
  import torch
141
 
142
  import qai_hub as hub
143
+ from qai_hub_models.models.aotgan import Model
144
 
145
  # Load the model
146
+ torch_model = Model.from_pretrained()
147
 
148
  # Device
149
  device = hub.Device("Samsung Galaxy S23")
150
 
151
+ # Trace model
152
+ input_shape = torch_model.get_input_spec()
153
+ sample_inputs = torch_model.sample_inputs()
154
+
155
+ pt_model = torch.jit.trace(torch_model, [torch.tensor(data[0]) for _, data in sample_inputs.items()])
156
+
157
+ # Compile model on a specific device
158
+ compile_job = hub.submit_compile_job(
159
+ model=pt_model,
160
+ device=device,
161
+ input_specs=torch_model.get_input_spec(),
162
+ )
163
+
164
+ # Get target model to run on-device
165
+ target_model = compile_job.get_target_model()
166
 
167
  ```
168