modelId
stringlengths 5
139
| author
stringlengths 2
42
| last_modified
timestamp[us, tz=UTC]date 2020-02-15 11:33:14
2025-06-05 12:28:32
| downloads
int64 0
223M
| likes
int64 0
11.7k
| library_name
stringclasses 468
values | tags
sequencelengths 1
4.05k
| pipeline_tag
stringclasses 54
values | createdAt
timestamp[us, tz=UTC]date 2022-03-02 23:29:04
2025-06-05 12:27:45
| card
stringlengths 11
1.01M
|
|---|---|---|---|---|---|---|---|---|---|
glasses/efficientnet_b3 | glasses | 2021-12-01T08:08:37Z | 2 | 0 | transformers | [
"transformers",
"pytorch",
"arxiv:1905.11946",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:05Z | # efficientnet_b3
Implementation of EfficientNet proposed in [EfficientNet: Rethinking
Model Scaling for Convolutional Neural
Networks](https://arxiv.org/abs/1905.11946)
The basic architecture is similar to MobileNetV2 as was computed by
using [Progressive Neural Architecture
Search](https://arxiv.org/abs/1905.11946) .
The following table shows the basic architecture
(EfficientNet-efficientnet\_b0):
Then, the architecture is scaled up from
[-efficientnet\_b0]{.title-ref} to [-efficientnet\_b7]{.title-ref}
using compound scaling.
``` python
EfficientNet.efficientnet_b0()
EfficientNet.efficientnet_b1()
EfficientNet.efficientnet_b2()
EfficientNet.efficientnet_b3()
EfficientNet.efficientnet_b4()
EfficientNet.efficientnet_b5()
EfficientNet.efficientnet_b6()
EfficientNet.efficientnet_b7()
EfficientNet.efficientnet_b8()
EfficientNet.efficientnet_l2()
```
Examples:
``` python
EfficientNet.efficientnet_b0(activation = nn.SELU)
# change number of classes (default is 1000 )
EfficientNet.efficientnet_b0(n_classes=100)
# pass a different block
EfficientNet.efficientnet_b0(block=...)
# store each feature
x = torch.rand((1, 3, 224, 224))
model = EfficientNet.efficientnet_b0()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
# [torch.Size([1, 32, 112, 112]), torch.Size([1, 24, 56, 56]), torch.Size([1, 40, 28, 28]), torch.Size([1, 80, 14, 14])]
```
|
glasses/efficientnet_b2 | glasses | 2021-12-01T08:08:06Z | 2 | 0 | transformers | [
"transformers",
"pytorch",
"arxiv:1905.11946",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:05Z | # efficientnet_b2
Implementation of EfficientNet proposed in [EfficientNet: Rethinking
Model Scaling for Convolutional Neural
Networks](https://arxiv.org/abs/1905.11946)
The basic architecture is similar to MobileNetV2 as was computed by
using [Progressive Neural Architecture
Search](https://arxiv.org/abs/1905.11946) .
The following table shows the basic architecture
(EfficientNet-efficientnet\_b0):
Then, the architecture is scaled up from
[-efficientnet\_b0]{.title-ref} to [-efficientnet\_b7]{.title-ref}
using compound scaling.
``` python
EfficientNet.efficientnet_b0()
EfficientNet.efficientnet_b1()
EfficientNet.efficientnet_b2()
EfficientNet.efficientnet_b3()
EfficientNet.efficientnet_b4()
EfficientNet.efficientnet_b5()
EfficientNet.efficientnet_b6()
EfficientNet.efficientnet_b7()
EfficientNet.efficientnet_b8()
EfficientNet.efficientnet_l2()
```
Examples:
``` python
EfficientNet.efficientnet_b0(activation = nn.SELU)
# change number of classes (default is 1000 )
EfficientNet.efficientnet_b0(n_classes=100)
# pass a different block
EfficientNet.efficientnet_b0(block=...)
# store each feature
x = torch.rand((1, 3, 224, 224))
model = EfficientNet.efficientnet_b0()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
# [torch.Size([1, 32, 112, 112]), torch.Size([1, 24, 56, 56]), torch.Size([1, 40, 28, 28]), torch.Size([1, 80, 14, 14])]
```
|
glasses/vgg13_bn | glasses | 2021-12-01T08:02:05Z | 1 | 0 | transformers | [
"transformers",
"pytorch",
"arxiv:1409.1556",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:05Z | # vgg13_bn
Implementation of VGG proposed in [Very Deep Convolutional Networks For
Large-Scale Image Recognition](https://arxiv.org/pdf/1409.1556.pdf)
``` python
VGG.vgg11()
VGG.vgg13()
VGG.vgg16()
VGG.vgg19()
VGG.vgg11_bn()
VGG.vgg13_bn()
VGG.vgg16_bn()
VGG.vgg19_bn()
```
Please be aware that the [bn]{.title-ref} models uses BatchNorm but
they are very old and people back then don\'t know the bias is
superfluous in a conv followed by a batchnorm.
Examples:
``` python
# change activation
VGG.vgg11(activation = nn.SELU)
# change number of classes (default is 1000 )
VGG.vgg11(n_classes=100)
# pass a different block
from nn.models.classification.senet import SENetBasicBlock
VGG.vgg11(block=SENetBasicBlock)
# store the features tensor after every block
```
|
glasses/vgg11 | glasses | 2021-12-01T07:53:25Z | 2 | 0 | transformers | [
"transformers",
"pytorch",
"arxiv:1409.1556",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:05Z | # vgg11
Implementation of VGG proposed in [Very Deep Convolutional Networks For
Large-Scale Image Recognition](https://arxiv.org/pdf/1409.1556.pdf)
``` python
VGG.vgg11()
VGG.vgg13()
VGG.vgg16()
VGG.vgg19()
VGG.vgg11_bn()
VGG.vgg13_bn()
VGG.vgg16_bn()
VGG.vgg19_bn()
```
Please be aware that the [bn]{.title-ref} models uses BatchNorm but
they are very old and people back then don\'t know the bias is
superfluous in a conv followed by a batchnorm.
Examples:
``` python
# change activation
VGG.vgg11(activation = nn.SELU)
# change number of classes (default is 1000 )
VGG.vgg11(n_classes=100)
# pass a different block
from nn.models.classification.senet import SENetBasicBlock
VGG.vgg11(block=SENetBasicBlock)
# store the features tensor after every block
```
|
glasses/densenet161 | glasses | 2021-12-01T07:50:20Z | 2 | 0 | transformers | [
"transformers",
"pytorch",
"arxiv:1608.06993",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:05Z | # densenet161
Implementation of DenseNet proposed in [Densely Connected Convolutional
Networks](https://arxiv.org/abs/1608.06993)
Create a default models
``` {.sourceCode .}
DenseNet.densenet121()
DenseNet.densenet161()
DenseNet.densenet169()
DenseNet.densenet201()
```
Examples:
``` {.sourceCode .}
# change activation
DenseNet.densenet121(activation = nn.SELU)
# change number of classes (default is 1000 )
DenseNet.densenet121(n_classes=100)
# pass a different block
DenseNet.densenet121(block=...)
# change the initial convolution
model = DenseNet.densenet121()
model.encoder.gate.conv1 = nn.Conv2d(3, 64, kernel_size=3)
# store each feature
x = torch.rand((1, 3, 224, 224))
model = DenseNet.densenet121()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
# [torch.Size([1, 128, 28, 28]), torch.Size([1, 256, 14, 14]), torch.Size([1, 512, 7, 7]), torch.Size([1, 1024, 7, 7])]
```
|
glasses/densenet201 | glasses | 2021-12-01T07:49:34Z | 4 | 0 | transformers | [
"transformers",
"pytorch",
"arxiv:1608.06993",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:05Z | # densenet201
Implementation of DenseNet proposed in [Densely Connected Convolutional
Networks](https://arxiv.org/abs/1608.06993)
Create a default models
``` {.sourceCode .}
DenseNet.densenet121()
DenseNet.densenet161()
DenseNet.densenet169()
DenseNet.densenet201()
```
Examples:
``` {.sourceCode .}
# change activation
DenseNet.densenet121(activation = nn.SELU)
# change number of classes (default is 1000 )
DenseNet.densenet121(n_classes=100)
# pass a different block
DenseNet.densenet121(block=...)
# change the initial convolution
model = DenseNet.densenet121()
model.encoder.gate.conv1 = nn.Conv2d(3, 64, kernel_size=3)
# store each feature
x = torch.rand((1, 3, 224, 224))
model = DenseNet.densenet121()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
# [torch.Size([1, 128, 28, 28]), torch.Size([1, 256, 14, 14]), torch.Size([1, 512, 7, 7]), torch.Size([1, 1024, 7, 7])]
```
|
glasses/regnety_008 | glasses | 2021-12-01T07:46:29Z | 4 | 0 | transformers | [
"transformers",
"pytorch",
"arxiv:2003.13678",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:05Z | # regnety_008
Implementation of RegNet proposed in [Designing Network Design
Spaces](https://arxiv.org/abs/2003.13678)
The main idea is to start with a high dimensional search space and
iteratively reduce the search space by empirically apply constrains
based on the best performing models sampled by the current search
space.
The resulting models are light, accurate, and faster than
EfficientNets (up to 5x times!)
For example, to go from $AnyNet_A$ to $AnyNet_B$ they fixed the
bottleneck ratio $b_i$ for all stage $i$. The following table shows
all the restrictions applied from one search space to the next one.
The paper is really well written and very interesting, I highly
recommended read it.
``` python
ResNet.regnetx_002()
ResNet.regnetx_004()
ResNet.regnetx_006()
ResNet.regnetx_008()
ResNet.regnetx_016()
ResNet.regnetx_040()
ResNet.regnetx_064()
ResNet.regnetx_080()
ResNet.regnetx_120()
ResNet.regnetx_160()
ResNet.regnetx_320()
# Y variants (with SE)
ResNet.regnety_002()
# ...
ResNet.regnetx_320()
You can easily customize your model
```
Examples:
``` python
# change activation
RegNet.regnetx_004(activation = nn.SELU)
# change number of classes (default is 1000 )
RegNet.regnetx_004(n_classes=100)
# pass a different block
RegNet.regnetx_004(block=RegNetYBotteneckBlock)
# change the steam
model = RegNet.regnetx_004(stem=ResNetStemC)
change shortcut
model = RegNet.regnetx_004(block=partial(RegNetYBotteneckBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = RegNet.regnetx_004()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 32, 112, 112]), torch.Size([1, 32, 56, 56]), torch.Size([1, 64, 28, 28]), torch.Size([1, 160, 14, 14])]
```
|
glasses/regnety_006 | glasses | 2021-12-01T07:46:05Z | 2 | 0 | transformers | [
"transformers",
"pytorch",
"arxiv:2003.13678",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:05Z | # regnety_006
Implementation of RegNet proposed in [Designing Network Design
Spaces](https://arxiv.org/abs/2003.13678)
The main idea is to start with a high dimensional search space and
iteratively reduce the search space by empirically apply constrains
based on the best performing models sampled by the current search
space.
The resulting models are light, accurate, and faster than
EfficientNets (up to 5x times!)
For example, to go from $AnyNet_A$ to $AnyNet_B$ they fixed the
bottleneck ratio $b_i$ for all stage $i$. The following table shows
all the restrictions applied from one search space to the next one.
The paper is really well written and very interesting, I highly
recommended read it.
``` python
ResNet.regnetx_002()
ResNet.regnetx_004()
ResNet.regnetx_006()
ResNet.regnetx_008()
ResNet.regnetx_016()
ResNet.regnetx_040()
ResNet.regnetx_064()
ResNet.regnetx_080()
ResNet.regnetx_120()
ResNet.regnetx_160()
ResNet.regnetx_320()
# Y variants (with SE)
ResNet.regnety_002()
# ...
ResNet.regnetx_320()
You can easily customize your model
```
Examples:
``` python
# change activation
RegNet.regnetx_004(activation = nn.SELU)
# change number of classes (default is 1000 )
RegNet.regnetx_004(n_classes=100)
# pass a different block
RegNet.regnetx_004(block=RegNetYBotteneckBlock)
# change the steam
model = RegNet.regnetx_004(stem=ResNetStemC)
change shortcut
model = RegNet.regnetx_004(block=partial(RegNetYBotteneckBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = RegNet.regnetx_004()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 32, 112, 112]), torch.Size([1, 32, 56, 56]), torch.Size([1, 64, 28, 28]), torch.Size([1, 160, 14, 14])]
```
|
mofawzy/argpt2-goodreads | mofawzy | 2021-12-01T06:55:41Z | 7 | 1 | transformers | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"generated_from_trainer",
"ar",
"dataset:LABR",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text-generation | 2022-03-02T23:29:05Z | ---
tags:
- generated_from_trainer
language: ar
datasets:
- LABR
widget:
- text: "كان الكاتب ممكن"
- text: "كتاب ممتاز ولكن"
- text: "رواية درامية جدا والافكار بسيطة"
model-index:
- name: argpt2-goodreads
results: []
---
# argpt2-goodreads
This model is a fine-tuned version of [gpt2-medium](https://huggingface.co/gpt2-medium) on an goodreads LABR dataset.
It achieves the following results on the evaluation set:
- Loss: 1.4389
## Model description
Generate sentences either positive/negative examples based on goodreads corpus in arabic language.
## Intended uses & limitations
the model fine-tuned on arabic language only with aspect to generate sentences such as reviews in order todo the same for other languages you need to fine-tune it in your own.
any harmful content generated by GPT2 should not be used in anywhere.
## Training and evaluation data
training and validation done on goodreads dataset LABR 80% for trainng and 20% for testing
## Usage
```
from transformers import AutoTokenizer, AutoModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained("mofawzy/argpt2-goodreads")
model = AutoModelForCausalLM.from_pretrained("mofawzy/argpt2-goodreads")
```
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- distributed_type: tpu
- num_devices: 8
- total_train_batch_size: 128
- total_eval_batch_size: 128
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 20.0
### Training results
- train_loss = 1.474
### Evaluation results
- eval_loss = 1.4389
### train metrics
- epoch = 20.0
- train_loss = 1.474
- train_runtime = 2:18:14.51
- train_samples = 108110
- train_samples_per_second = 260.678
- train_steps_per_second = 2.037
### eval metrics
- epoch = 20.0
- eval_loss = 1.4389
- eval_runtime = 0:04:37.01
- eval_samples = 27329
- eval_samples_per_second = 98.655
- eval_steps_per_second = 0.773
- perplexity = 4.2162
### Framework versions
- Transformers 4.13.0.dev0
- Pytorch 1.10.0+cu102
- Datasets 1.16.1
- Tokenizers 0.10.3
|
MMG/bert-base-spanish-wwm-cased-finetuned-sqac | MMG | 2021-12-01T06:13:29Z | 34 | 0 | transformers | [
"transformers",
"pytorch",
"tensorboard",
"bert",
"question-answering",
"generated_from_trainer",
"es",
"dataset:sqac",
"endpoints_compatible",
"region:us"
] | question-answering | 2022-03-02T23:29:04Z | ---
tags:
- generated_from_trainer
datasets:
- sqac
model-index:
- name: bert-base-spanish-wwm-cased-finetuned-sqac
results: []
language:
- es
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# bert-base-spanish-wwm-cased-finetuned-sqac
This model is a fine-tuned version of [dccuchile/bert-base-spanish-wwm-cased](https://huggingface.co/dccuchile/bert-base-spanish-wwm-cased) on the sqac dataset.
It achieves the following results on the evaluation set:
{'exact_match': 62.017167, 'f1': 79.452767}
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 3
### Training results
| Training Loss | Epoch | Step | Validation Loss |
|:-------------:|:-----:|:----:|:---------------:|
| 1.1335 | 1.0 | 1230 | 0.9346 |
| 0.6794 | 2.0 | 2460 | 0.8634 |
| 0.3992 | 3.0 | 3690 | 0.9662 |
### Framework versions
- Transformers 4.12.5
- Pytorch 1.10.0+cu111
- Datasets 1.16.1
- Tokenizers 0.10.3
|
ykliu1892/translation-en-pt-t5-finetuned-Duolingo | ykliu1892 | 2021-12-01T04:58:54Z | 4 | 0 | transformers | [
"transformers",
"pytorch",
"tensorboard",
"t5",
"text2text-generation",
"generated_from_trainer",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text2text-generation | 2022-03-02T23:29:05Z | ---
tags:
- generated_from_trainer
metrics:
- bleu
model-index:
- name: translation-en-pt-t5-finetuned-Duolingo
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# translation-en-pt-t5-finetuned-Duolingo
This model was trained from scratch on an unknown dataset.
It achieves the following results on the evaluation set:
- Loss: 0.7362
- Bleu: 39.4725
- Gen Len: 9.002
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 32
- eval_batch_size: 32
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 1
- mixed_precision_training: Native AMP
### Training results
| Training Loss | Epoch | Step | Validation Loss | Bleu | Gen Len |
|:-------------:|:-----:|:-----:|:---------------:|:-------:|:-------:|
| 0.5429 | 0.24 | 9000 | 0.7461 | 39.4744 | 9.0 |
| 0.5302 | 0.48 | 18000 | 0.7431 | 39.7559 | 8.97 |
| 0.5309 | 0.72 | 27000 | 0.7388 | 39.6751 | 8.998 |
| 0.5336 | 0.96 | 36000 | 0.7362 | 39.4725 | 9.002 |
### Framework versions
- Transformers 4.12.5
- Pytorch 1.10.0+cu111
- Datasets 1.16.1
- Tokenizers 0.10.3
|
kaporter/bert-base-uncased-finetuned-squad | kaporter | 2021-11-30T22:42:17Z | 267 | 0 | transformers | [
"transformers",
"pytorch",
"tensorboard",
"bert",
"question-answering",
"generated_from_trainer",
"dataset:squad",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | question-answering | 2022-03-02T23:29:05Z | ---
license: apache-2.0
tags:
- generated_from_trainer
datasets:
- squad
model_index:
- name: bert-base-uncased-finetuned-squad
results:
- task:
name: Question Answering
type: question-answering
dataset:
name: squad
type: squad
args: plain_text
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# bert-base-uncased-finetuned-squad
This model is a fine-tuned version of [bert-base-uncased](https://huggingface.co/bert-base-uncased) on the squad dataset.
It achieves the following results on the evaluation set:
- Loss: 1.0725
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 3
### Training results
| Training Loss | Epoch | Step | Validation Loss |
|:-------------:|:-----:|:-----:|:---------------:|
| 1.0749 | 1.0 | 5533 | 1.0167 |
| 0.7851 | 2.0 | 11066 | 1.0299 |
| 0.6067 | 3.0 | 16599 | 1.0725 |
### Framework versions
- Transformers 4.8.1
- Pytorch 1.8.1
- Datasets 1.16.1
- Tokenizers 0.10.1
|
mmcquade11-test/reuters-summarization | mmcquade11-test | 2021-11-30T21:43:51Z | 4 | 0 | transformers | [
"transformers",
"pytorch",
"pegasus",
"text2text-generation",
"autonlp",
"en",
"dataset:mmcquade11/autonlp-data-reuters-summarization",
"co2_eq_emissions",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text2text-generation | 2022-03-02T23:29:05Z | ---
tags: autonlp
language: en
widget:
- text: "I love AutoNLP 🤗"
datasets:
- mmcquade11/autonlp-data-reuters-summarization
co2_eq_emissions: 286.4350821612984
---
This is an autoNLP model I trained on Reuters dataset
# Model Trained Using AutoNLP
- Problem type: Summarization
- Model ID: 34018133
- CO2 Emissions (in grams): 286.4350821612984
## Validation Metrics
- Loss: 1.1805976629257202
- Rouge1: 55.4013
- Rouge2: 30.8004
- RougeL: 52.57
- RougeLsum: 52.6103
- Gen Len: 15.3458
## Usage
You can use cURL to access this model:
```
$ curl -X POST -H "Authorization: Bearer YOUR_HUGGINGFACE_API_KEY" -H "Content-Type: application/json" -d '{"inputs": "I love AutoNLP"}' https://api-inference.huggingface.co/mmcquade11/autonlp-reuters-summarization-34018133
``` |
glasses/eca_resnet26t | glasses | 2021-11-30T20:21:22Z | 31 | 0 | transformers | [
"transformers",
"pytorch",
"image-classification",
"dataset:imagenet",
"arxiv:1512.03385",
"arxiv:1812.01187",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | image-classification | 2022-03-02T23:29:05Z | ---
license: apache-2.0
tags:
- image-classification
datasets:
- imagenet
---
# eca_resnet26t
Implementation of ResNet proposed in [Deep Residual Learning for Image
Recognition](https://arxiv.org/abs/1512.03385)
``` python
ResNet.resnet18()
ResNet.resnet26()
ResNet.resnet34()
ResNet.resnet50()
ResNet.resnet101()
ResNet.resnet152()
ResNet.resnet200()
Variants (d) proposed in `Bag of Tricks for Image Classification with Convolutional Neural Networks <https://arxiv.org/pdf/1812.01187.pdf`_
ResNet.resnet26d()
ResNet.resnet34d()
ResNet.resnet50d()
# You can construct your own one by chaning `stem` and `block`
resnet101d = ResNet.resnet101(stem=ResNetStemC, block=partial(ResNetBottleneckBlock, shortcut=ResNetShorcutD))
```
Examples:
``` python
# change activation
ResNet.resnet18(activation = nn.SELU)
# change number of classes (default is 1000 )
ResNet.resnet18(n_classes=100)
# pass a different block
ResNet.resnet18(block=SENetBasicBlock)
# change the steam
model = ResNet.resnet18(stem=ResNetStemC)
change shortcut
model = ResNet.resnet18(block=partial(ResNetBasicBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = ResNet.resnet18()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 64, 112, 112]), torch.Size([1, 64, 56, 56]), torch.Size([1, 128, 28, 28]), torch.Size([1, 256, 14, 14])]
```
|
glasses/resnext101_32x8d | glasses | 2021-11-30T20:15:04Z | 1 | 0 | transformers | [
"transformers",
"pytorch",
"arxiv:1611.05431",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:05Z | # resnext101_32x8d
Implementation of ResNetXt proposed in [\"Aggregated Residual
Transformation for Deep Neural
Networks\"](https://arxiv.org/pdf/1611.05431.pdf)
Create a default model
``` python
ResNetXt.resnext50_32x4d()
ResNetXt.resnext101_32x8d()
# create a resnetxt18_32x4d
ResNetXt.resnet18(block=ResNetXtBottleNeckBlock, groups=32, base_width=4)
```
Examples:
: ``` python
# change activation
ResNetXt.resnext50_32x4d(activation = nn.SELU)
# change number of classes (default is 1000 )
ResNetXt.resnext50_32x4d(n_classes=100)
# pass a different block
ResNetXt.resnext50_32x4d(block=SENetBasicBlock)
# change the initial convolution
model = ResNetXt.resnext50_32x4d
model.encoder.gate.conv1 = nn.Conv2d(3, 64, kernel_size=3)
# store each feature
x = torch.rand((1, 3, 224, 224))
model = ResNetXt.resnext50_32x4d()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 64, 112, 112]), torch.Size([1, 64, 56, 56]), torch.Size([1, 128, 28, 28]), torch.Size([1, 256, 14, 14])]
```
|
glasses/resnet152 | glasses | 2021-11-30T20:12:19Z | 30 | 0 | transformers | [
"transformers",
"pytorch",
"image-classification",
"dataset:imagenet",
"arxiv:1512.03385",
"arxiv:1812.01187",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | image-classification | 2022-03-02T23:29:05Z | ---
license: apache-2.0
tags:
- image-classification
datasets:
- imagenet
---
# resnet152
Implementation of ResNet proposed in [Deep Residual Learning for Image
Recognition](https://arxiv.org/abs/1512.03385)
``` python
ResNet.resnet18()
ResNet.resnet26()
ResNet.resnet34()
ResNet.resnet50()
ResNet.resnet101()
ResNet.resnet152()
ResNet.resnet200()
Variants (d) proposed in `Bag of Tricks for Image Classification with Convolutional Neural Networks <https://arxiv.org/pdf/1812.01187.pdf`_
ResNet.resnet26d()
ResNet.resnet34d()
ResNet.resnet50d()
# You can construct your own one by chaning `stem` and `block`
resnet101d = ResNet.resnet101(stem=ResNetStemC, block=partial(ResNetBottleneckBlock, shortcut=ResNetShorcutD))
```
Examples:
``` python
# change activation
ResNet.resnet18(activation = nn.SELU)
# change number of classes (default is 1000 )
ResNet.resnet18(n_classes=100)
# pass a different block
ResNet.resnet18(block=SENetBasicBlock)
# change the steam
model = ResNet.resnet18(stem=ResNetStemC)
change shortcut
model = ResNet.resnet18(block=partial(ResNetBasicBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = ResNet.resnet18()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 64, 112, 112]), torch.Size([1, 64, 56, 56]), torch.Size([1, 128, 28, 28]), torch.Size([1, 256, 14, 14])]
```
|
glasses/resnet50d | glasses | 2021-11-30T20:10:20Z | 29 | 0 | transformers | [
"transformers",
"pytorch",
"image-classification",
"dataset:imagenet",
"arxiv:1512.03385",
"arxiv:1812.01187",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | image-classification | 2022-03-02T23:29:05Z | ---
license: apache-2.0
tags:
- image-classification
datasets:
- imagenet
---
# resnet50d
Implementation of ResNet proposed in [Deep Residual Learning for Image
Recognition](https://arxiv.org/abs/1512.03385)
``` python
ResNet.resnet18()
ResNet.resnet26()
ResNet.resnet34()
ResNet.resnet50()
ResNet.resnet101()
ResNet.resnet152()
ResNet.resnet200()
Variants (d) proposed in `Bag of Tricks for Image Classification with Convolutional Neural Networks <https://arxiv.org/pdf/1812.01187.pdf`_
ResNet.resnet26d()
ResNet.resnet34d()
ResNet.resnet50d()
# You can construct your own one by chaning `stem` and `block`
resnet101d = ResNet.resnet101(stem=ResNetStemC, block=partial(ResNetBottleneckBlock, shortcut=ResNetShorcutD))
```
Examples:
``` python
# change activation
ResNet.resnet18(activation = nn.SELU)
# change number of classes (default is 1000 )
ResNet.resnet18(n_classes=100)
# pass a different block
ResNet.resnet18(block=SENetBasicBlock)
# change the steam
model = ResNet.resnet18(stem=ResNetStemC)
change shortcut
model = ResNet.resnet18(block=partial(ResNetBasicBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = ResNet.resnet18()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 64, 112, 112]), torch.Size([1, 64, 56, 56]), torch.Size([1, 128, 28, 28]), torch.Size([1, 256, 14, 14])]
```
|
glasses/resnet50 | glasses | 2021-11-30T20:09:35Z | 29 | 0 | transformers | [
"transformers",
"pytorch",
"image-classification",
"dataset:imagenet",
"arxiv:1512.03385",
"arxiv:1812.01187",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | image-classification | 2022-03-02T23:29:05Z | ---
license: apache-2.0
tags:
- image-classification
datasets:
- imagenet
---
# resnet50
Implementation of ResNet proposed in [Deep Residual Learning for Image
Recognition](https://arxiv.org/abs/1512.03385)
``` python
ResNet.resnet18()
ResNet.resnet26()
ResNet.resnet34()
ResNet.resnet50()
ResNet.resnet101()
ResNet.resnet152()
ResNet.resnet200()
Variants (d) proposed in `Bag of Tricks for Image Classification with Convolutional Neural Networks <https://arxiv.org/pdf/1812.01187.pdf`_
ResNet.resnet26d()
ResNet.resnet34d()
ResNet.resnet50d()
# You can construct your own one by chaning `stem` and `block`
resnet101d = ResNet.resnet101(stem=ResNetStemC, block=partial(ResNetBottleneckBlock, shortcut=ResNetShorcutD))
```
Examples:
``` python
# change activation
ResNet.resnet18(activation = nn.SELU)
# change number of classes (default is 1000 )
ResNet.resnet18(n_classes=100)
# pass a different block
ResNet.resnet18(block=SENetBasicBlock)
# change the steam
model = ResNet.resnet18(stem=ResNetStemC)
change shortcut
model = ResNet.resnet18(block=partial(ResNetBasicBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = ResNet.resnet18()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 64, 112, 112]), torch.Size([1, 64, 56, 56]), torch.Size([1, 128, 28, 28]), torch.Size([1, 256, 14, 14])]
```
|
glasses/resnet34d | glasses | 2021-11-30T20:08:51Z | 36 | 0 | transformers | [
"transformers",
"pytorch",
"image-classification",
"dataset:imagenet",
"arxiv:1512.03385",
"arxiv:1812.01187",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | image-classification | 2022-03-02T23:29:05Z | ---
license: apache-2.0
tags:
- image-classification
datasets:
- imagenet
---
# resnet34d
Implementation of ResNet proposed in [Deep Residual Learning for Image
Recognition](https://arxiv.org/abs/1512.03385)
``` python
ResNet.resnet18()
ResNet.resnet26()
ResNet.resnet34()
ResNet.resnet50()
ResNet.resnet101()
ResNet.resnet152()
ResNet.resnet200()
Variants (d) proposed in `Bag of Tricks for Image Classification with Convolutional Neural Networks <https://arxiv.org/pdf/1812.01187.pdf`_
ResNet.resnet26d()
ResNet.resnet34d()
ResNet.resnet50d()
# You can construct your own one by chaning `stem` and `block`
resnet101d = ResNet.resnet101(stem=ResNetStemC, block=partial(ResNetBottleneckBlock, shortcut=ResNetShorcutD))
```
Examples:
``` python
# change activation
ResNet.resnet18(activation = nn.SELU)
# change number of classes (default is 1000 )
ResNet.resnet18(n_classes=100)
# pass a different block
ResNet.resnet18(block=SENetBasicBlock)
# change the steam
model = ResNet.resnet18(stem=ResNetStemC)
change shortcut
model = ResNet.resnet18(block=partial(ResNetBasicBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = ResNet.resnet18()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 64, 112, 112]), torch.Size([1, 64, 56, 56]), torch.Size([1, 128, 28, 28]), torch.Size([1, 256, 14, 14])]
```
|
glasses/resnet34 | glasses | 2021-11-30T20:08:12Z | 33 | 0 | transformers | [
"transformers",
"pytorch",
"image-classification",
"dataset:imagenet",
"arxiv:1512.03385",
"arxiv:1812.01187",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | image-classification | 2022-03-02T23:29:05Z | ---
license: apache-2.0
tags:
- image-classification
datasets:
- imagenet
---
# resnet34
Implementation of ResNet proposed in [Deep Residual Learning for Image
Recognition](https://arxiv.org/abs/1512.03385)
``` python
ResNet.resnet18()
ResNet.resnet26()
ResNet.resnet34()
ResNet.resnet50()
ResNet.resnet101()
ResNet.resnet152()
ResNet.resnet200()
Variants (d) proposed in `Bag of Tricks for Image Classification with Convolutional Neural Networks <https://arxiv.org/pdf/1812.01187.pdf`_
ResNet.resnet26d()
ResNet.resnet34d()
ResNet.resnet50d()
# You can construct your own one by chaning `stem` and `block`
resnet101d = ResNet.resnet101(stem=ResNetStemC, block=partial(ResNetBottleneckBlock, shortcut=ResNetShorcutD))
```
Examples:
``` python
# change activation
ResNet.resnet18(activation = nn.SELU)
# change number of classes (default is 1000 )
ResNet.resnet18(n_classes=100)
# pass a different block
ResNet.resnet18(block=SENetBasicBlock)
# change the steam
model = ResNet.resnet18(stem=ResNetStemC)
change shortcut
model = ResNet.resnet18(block=partial(ResNetBasicBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = ResNet.resnet18()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 64, 112, 112]), torch.Size([1, 64, 56, 56]), torch.Size([1, 128, 28, 28]), torch.Size([1, 256, 14, 14])]
```
|
glasses/resnet18 | glasses | 2021-11-30T20:06:28Z | 37 | 0 | transformers | [
"transformers",
"pytorch",
"image-classification",
"dataset:imagenet",
"arxiv:1512.03385",
"arxiv:1812.01187",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | image-classification | 2022-03-02T23:29:05Z | ---
license: apache-2.0
tags:
- image-classification
datasets:
- imagenet
---
# resnet18
Implementation of ResNet proposed in [Deep Residual Learning for Image
Recognition](https://arxiv.org/abs/1512.03385)
``` python
ResNet.resnet18()
ResNet.resnet26()
ResNet.resnet34()
ResNet.resnet50()
ResNet.resnet101()
ResNet.resnet152()
ResNet.resnet200()
Variants (d) proposed in `Bag of Tricks for Image Classification with Convolutional Neural Networks <https://arxiv.org/pdf/1812.01187.pdf`_
ResNet.resnet26d()
ResNet.resnet34d()
ResNet.resnet50d()
# You can construct your own one by chaning `stem` and `block`
resnet101d = ResNet.resnet101(stem=ResNetStemC, block=partial(ResNetBottleneckBlock, shortcut=ResNetShorcutD))
```
Examples:
``` python
# change activation
ResNet.resnet18(activation = nn.SELU)
# change number of classes (default is 1000 )
ResNet.resnet18(n_classes=100)
# pass a different block
ResNet.resnet18(block=SENetBasicBlock)
# change the steam
model = ResNet.resnet18(stem=ResNetStemC)
change shortcut
model = ResNet.resnet18(block=partial(ResNetBasicBlock, shortcut=ResNetShorcutD))
# store each feature
x = torch.rand((1, 3, 224, 224))
# get features
model = ResNet.resnet18()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
#[torch.Size([1, 64, 112, 112]), torch.Size([1, 64, 56, 56]), torch.Size([1, 128, 28, 28]), torch.Size([1, 256, 14, 14])]
```
|
ffsouza/tiny-mbart-length-96-learning_rate-2e-05-weight_decay-0.005-finetuned-en-to-ro | ffsouza | 2021-11-30T19:57:36Z | 26 | 0 | transformers | [
"transformers",
"pytorch",
"tensorboard",
"mbart",
"text2text-generation",
"generated_from_trainer",
"dataset:wmt16_en_ro_pre_processed",
"model-index",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text2text-generation | 2022-03-02T23:29:05Z | ---
tags:
- generated_from_trainer
datasets:
- wmt16_en_ro_pre_processed
metrics:
- bleu
model-index:
- name: tiny-mbart-length-96-learning_rate-2e-05-weight_decay-0.005-finetuned-en-to-ro
results:
- task:
name: Sequence-to-sequence Language Modeling
type: text2text-generation
dataset:
name: wmt16_en_ro_pre_processed
type: wmt16_en_ro_pre_processed
args: enro
metrics:
- name: Bleu
type: bleu
value: 0.0
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# tiny-mbart-length-96-learning_rate-2e-05-weight_decay-0.005-finetuned-en-to-ro
This model is a fine-tuned version of [sshleifer/tiny-mbart](https://huggingface.co/sshleifer/tiny-mbart) on the wmt16_en_ro_pre_processed dataset.
It achieves the following results on the evaluation set:
- Loss: 8.5983
- Bleu: 0.0
- Gen Len: 20.0
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 1
- mixed_precision_training: Native AMP
### Training results
| Training Loss | Epoch | Step | Validation Loss | Bleu | Gen Len |
|:-------------:|:-----:|:-----:|:---------------:|:----:|:-------:|
| 8.3753 | 1.0 | 76290 | 8.5983 | 0.0 | 20.0 |
### Framework versions
- Transformers 4.12.5
- Pytorch 1.10.0+cu102
- Datasets 1.15.1
- Tokenizers 0.10.3
|
tyoyo/t5-base-TEDxJP-1body-10context | tyoyo | 2021-11-30T19:40:13Z | 10 | 0 | transformers | [
"transformers",
"pytorch",
"tensorboard",
"t5",
"text2text-generation",
"generated_from_trainer",
"dataset:te_dx_jp",
"license:cc-by-sa-4.0",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text2text-generation | 2022-03-02T23:29:05Z | ---
license: cc-by-sa-4.0
tags:
- generated_from_trainer
datasets:
- te_dx_jp
model-index:
- name: t5-base-TEDxJP-1body-10context
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# t5-base-TEDxJP-1body-10context
This model is a fine-tuned version of [sonoisa/t5-base-japanese](https://huggingface.co/sonoisa/t5-base-japanese) on the te_dx_jp dataset.
It achieves the following results on the evaluation set:
- Loss: 0.3833
- Wer: 0.1983
- Mer: 0.1900
- Wil: 0.2778
- Wip: 0.7222
- Hits: 56229
- Substitutions: 6686
- Deletions: 3593
- Insertions: 2909
- Cer: 0.1823
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 0.0001
- train_batch_size: 64
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 10
### Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Mer | Wil | Wip | Hits | Substitutions | Deletions | Insertions | Cer |
|:-------------:|:-----:|:----:|:---------------:|:------:|:------:|:------:|:------:|:-----:|:-------------:|:---------:|:----------:|:------:|
| 0.5641 | 1.0 | 746 | 0.4426 | 0.2336 | 0.2212 | 0.3143 | 0.6857 | 54711 | 7183 | 4614 | 3742 | 0.2238 |
| 0.4867 | 2.0 | 1492 | 0.4017 | 0.2045 | 0.1972 | 0.2863 | 0.7137 | 55378 | 6764 | 4366 | 2470 | 0.1853 |
| 0.4257 | 3.0 | 2238 | 0.3831 | 0.2008 | 0.1933 | 0.2826 | 0.7174 | 55715 | 6788 | 4005 | 2560 | 0.1784 |
| 0.4038 | 4.0 | 2984 | 0.3797 | 0.1963 | 0.1890 | 0.2776 | 0.7224 | 56028 | 6731 | 3749 | 2578 | 0.1748 |
| 0.3817 | 5.0 | 3730 | 0.3769 | 0.1944 | 0.1877 | 0.2758 | 0.7242 | 55926 | 6663 | 3919 | 2345 | 0.1730 |
| 0.3467 | 6.0 | 4476 | 0.3806 | 0.2111 | 0.2002 | 0.2876 | 0.7124 | 56082 | 6688 | 3738 | 3616 | 0.1916 |
| 0.3361 | 7.0 | 5222 | 0.3797 | 0.1977 | 0.1897 | 0.2780 | 0.7220 | 56173 | 6721 | 3614 | 2816 | 0.1785 |
| 0.3107 | 8.0 | 5968 | 0.3814 | 0.1993 | 0.1910 | 0.2792 | 0.7208 | 56167 | 6720 | 3621 | 2916 | 0.1839 |
| 0.3141 | 9.0 | 6714 | 0.3820 | 0.1991 | 0.1907 | 0.2787 | 0.7213 | 56201 | 6709 | 3598 | 2933 | 0.1859 |
| 0.3122 | 10.0 | 7460 | 0.3833 | 0.1983 | 0.1900 | 0.2778 | 0.7222 | 56229 | 6686 | 3593 | 2909 | 0.1823 |
### Framework versions
- Transformers 4.12.5
- Pytorch 1.10.0+cu102
- Datasets 1.15.1
- Tokenizers 0.10.3
|
kenlevine/distilbert-base-uncased-finetuned-squad | kenlevine | 2021-11-30T18:04:35Z | 6 | 0 | transformers | [
"transformers",
"pytorch",
"tensorboard",
"distilbert",
"question-answering",
"generated_from_trainer",
"dataset:squad",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | question-answering | 2022-03-02T23:29:05Z | ---
license: apache-2.0
tags:
- generated_from_trainer
datasets:
- squad
model-index:
- name: distilbert-base-uncased-finetuned-squad
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# distilbert-base-uncased-finetuned-squad
This model is a fine-tuned version of [distilbert-base-uncased](https://huggingface.co/distilbert-base-uncased) on the squad dataset.
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 3
### Framework versions
- Transformers 4.12.5
- Pytorch 1.10.0+cu111
- Datasets 1.16.1
- Tokenizers 0.10.3
|
Raphaelg9/distilbert-base-uncased-finetuned-squad | Raphaelg9 | 2021-11-30T17:30:54Z | 4 | 0 | transformers | [
"transformers",
"pytorch",
"tensorboard",
"distilbert",
"question-answering",
"generated_from_trainer",
"dataset:squad_v2",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | question-answering | 2022-03-02T23:29:04Z | ---
license: apache-2.0
tags:
- generated_from_trainer
datasets:
- squad_v2
model-index:
- name: distilbert-base-uncased-finetuned-squad
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# distilbert-base-uncased-finetuned-squad
This model is a fine-tuned version of [distilbert-base-uncased](https://huggingface.co/distilbert-base-uncased) on the squad_v2 dataset.
It achieves the following results on the evaluation set:
- Loss: 2.1323
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 3
### Training results
| Training Loss | Epoch | Step | Validation Loss |
|:-------------:|:-----:|:----:|:---------------:|
| 2.8535 | 1.0 | 661 | 2.0684 |
| 1.5385 | 2.0 | 1322 | 2.0954 |
| 1.2312 | 3.0 | 1983 | 2.1323 |
### Framework versions
- Transformers 4.12.5
- Pytorch 1.10.0+cu111
- Datasets 1.16.1
- Tokenizers 0.10.3
|
huggingtweets/hel_ql-shahdashrf_-sinnerslayerr-witheredstrings | huggingtweets | 2021-11-30T15:40:26Z | 3 | 0 | transformers | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"huggingtweets",
"en",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text-generation | 2022-03-02T23:29:05Z | ---
language: en
thumbnail: https://www.huggingtweets.com/hel_ql-shahdashrf_-sinnerslayerr-witheredstrings/1638286821619/predictions.png
tags:
- huggingtweets
widget:
- text: "My dream is"
---
<div class="inline-flex flex-col" style="line-height: 1.5;">
<div class="flex">
<div
style="display:inherit; margin-left: 4px; margin-right: 4px; width: 92px; height:92px; border-radius: 50%; background-size: cover; background-image: url('https://pbs.twimg.com/profile_images/1449201367080386564/GllCx8JB_400x400.jpg')">
</div>
<div
style="display:inherit; margin-left: 4px; margin-right: 4px; width: 92px; height:92px; border-radius: 50%; background-size: cover; background-image: url('https://pbs.twimg.com/profile_images/1461790972392656898/e1248oRI_400x400.jpg')">
</div>
<div
style="display:inherit; margin-left: 4px; margin-right: 4px; width: 92px; height:92px; border-radius: 50%; background-size: cover; background-image: url('https://pbs.twimg.com/profile_images/1457045233783701504/fnjAg6lH_400x400.jpg')">
</div>
</div>
<div style="text-align: center; margin-top: 3px; font-size: 16px; font-weight: 800">🤖 AI CYBORG 🤖</div>
<div style="text-align: center; font-size: 16px; font-weight: 800">Sinner & Hσɳҽყ & Anthropos & VacuumF</div>
<div style="text-align: center; font-size: 14px;">@hel_ql-shahdashrf_-sinnerslayerr-witheredstrings</div>
</div>
I was made with [huggingtweets](https://github.com/borisdayma/huggingtweets).
Create your own bot based on your favorite user with [the demo](https://colab.research.google.com/github/borisdayma/huggingtweets/blob/master/huggingtweets-demo.ipynb)!
## How does it work?
The model uses the following pipeline.
To understand how the model was developed, check the [W&B report](https://wandb.ai/wandb/huggingtweets/reports/HuggingTweets-Train-a-Model-to-Generate-Tweets--VmlldzoxMTY5MjI).
## Training data
The model was trained on tweets from Sinner & Hσɳҽყ & Anthropos & VacuumF.
| Data | Sinner | Hσɳҽყ | Anthropos | VacuumF |
| --- | --- | --- | --- | --- |
| Tweets downloaded | 403 | 3240 | 1088 | 379 |
| Retweets | 296 | 135 | 376 | 1 |
| Short tweets | 3 | 734 | 77 | 12 |
| Tweets kept | 104 | 2371 | 635 | 366 |
[Explore the data](https://wandb.ai/wandb/huggingtweets/runs/2fhsvt3r/artifacts), which is tracked with [W&B artifacts](https://docs.wandb.com/artifacts) at every step of the pipeline.
## Training procedure
The model is based on a pre-trained [GPT-2](https://huggingface.co/gpt2) which is fine-tuned on @hel_ql-shahdashrf_-sinnerslayerr-witheredstrings's tweets.
Hyperparameters and metrics are recorded in the [W&B training run](https://wandb.ai/wandb/huggingtweets/runs/2kjvpfsa) for full transparency and reproducibility.
At the end of training, [the final model](https://wandb.ai/wandb/huggingtweets/runs/2kjvpfsa/artifacts) is logged and versioned.
## How to use
You can use this model directly with a pipeline for text generation:
```python
from transformers import pipeline
generator = pipeline('text-generation',
model='huggingtweets/hel_ql-shahdashrf_-sinnerslayerr-witheredstrings')
generator("My dream is", num_return_sequences=5)
```
## Limitations and bias
The model suffers from [the same limitations and bias as GPT-2](https://huggingface.co/gpt2#limitations-and-bias).
In addition, the data present in the user's tweets further affects the text generated by the model.
## About
*Built by Boris Dayma*
[](https://twitter.com/intent/follow?screen_name=borisdayma)
For more details, visit the project repository.
[](https://github.com/borisdayma/huggingtweets)
|
ptro/model1_test | ptro | 2021-11-30T15:25:05Z | 6 | 1 | transformers | [
"transformers",
"pytorch",
"tensorboard",
"bert",
"text-classification",
"generated_from_trainer",
"license:cc-by-sa-4.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text-classification | 2022-03-02T23:29:05Z | ---
license: cc-by-sa-4.0
tags:
- generated_from_trainer
metrics:
- accuracy
- f1
model-index:
- name: model1_test
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# model1_test
This model is a fine-tuned version of [DaNLP/da-bert-hatespeech-detection](https://huggingface.co/DaNLP/da-bert-hatespeech-detection) on an unknown dataset.
It achieves the following results on the evaluation set:
- Loss: 0.1816
- Accuracy: 0.9667
- F1: 0.3548
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 3.0
### Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy | F1 |
|:-------------:|:-----:|:----:|:---------------:|:--------:|:------:|
| No log | 1.0 | 150 | 0.1128 | 0.9667 | 0.2 |
| No log | 2.0 | 300 | 0.1666 | 0.9684 | 0.2963 |
| No log | 3.0 | 450 | 0.1816 | 0.9667 | 0.3548 |
### Framework versions
- Transformers 4.12.5
- Pytorch 1.10.0+cu111
- Datasets 1.16.1
- Tokenizers 0.10.3
|
tyoyo/t5-base-TEDxJP-1body-5context | tyoyo | 2021-11-30T13:49:54Z | 5 | 0 | transformers | [
"transformers",
"pytorch",
"tensorboard",
"t5",
"text2text-generation",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text2text-generation | 2022-03-02T23:29:05Z | Epoch Training Loss Validation Loss Wer Mer Wil Wip Hits Substitutions Deletions Insertions Cer
1 0.572400 0.447836 0.262284 0.241764 0.333088 0.666912 54709 7126 4673 5645 0.242417
2 0.492700 0.400297 0.203600 0.196446 0.285798 0.714202 55389 6777 4342 2422 0.183740
3 0.429200 0.385705 0.201179 0.193641 0.282458 0.717542 55717 6745 4046 2589 0.179833
4 0.408700 0.383085 0.198277 0.190817 0.280919 0.719081 55921 6867 3720 2600 0.177468
5 0.386100 0.381157 0.192488 0.186279 0.274890 0.725110 55923 6709 3876 2217 0.171644
6 0.353400 0.380517 0.193315 0.186615 0.275510 0.724490 56039 6747 3722 2388 0.170799
7 0.346100 0.379445 0.194713 0.187616 0.276780 0.723220 56074 6780 3654 2516 0.171347
8 0.314700 0.383521 0.196022 0.188486 0.277974 0.722026 56130 6820 3558 2659 0.179184
|
abhishek/autonlp-bbc-roberta-37249301 | abhishek | 2021-11-30T13:35:38Z | 11 | 0 | transformers | [
"transformers",
"pytorch",
"roberta",
"text-classification",
"autonlp",
"unk",
"dataset:abhishek/autonlp-data-bbc-roberta",
"co2_eq_emissions",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text-classification | 2022-03-02T23:29:05Z | ---
tags: autonlp
language: unk
widget:
- text: "I love AutoNLP 🤗"
datasets:
- abhishek/autonlp-data-bbc-roberta
co2_eq_emissions: 1.9859980179658823
---
# Model Trained Using AutoNLP
- Problem type: Multi-class Classification
- Model ID: 37249301
- CO2 Emissions (in grams): 1.9859980179658823
## Validation Metrics
- Loss: 0.06406362354755402
- Accuracy: 0.9833887043189369
- Macro F1: 0.9832763664701248
- Micro F1: 0.9833887043189369
- Weighted F1: 0.9833288528828136
- Macro Precision: 0.9847257743677181
- Micro Precision: 0.9833887043189369
- Weighted Precision: 0.9835392869652073
- Macro Recall: 0.982101705176067
- Micro Recall: 0.9833887043189369
- Weighted Recall: 0.9833887043189369
## Usage
You can use cURL to access this model:
```
$ curl -X POST -H "Authorization: Bearer YOUR_API_KEY" -H "Content-Type: application/json" -d '{"inputs": "I love AutoNLP"}' https://api-inference.huggingface.co/models/abhishek/autonlp-bbc-roberta-37249301
```
Or Python API:
```
from transformers import AutoModelForSequenceClassification, AutoTokenizer
model = AutoModelForSequenceClassification.from_pretrained("abhishek/autonlp-bbc-roberta-37249301", use_auth_token=True)
tokenizer = AutoTokenizer.from_pretrained("abhishek/autonlp-bbc-roberta-37249301", use_auth_token=True)
inputs = tokenizer("I love AutoNLP", return_tensors="pt")
outputs = model(**inputs)
``` |
abhishek/autonlp-bbc-news-classification-37229289 | abhishek | 2021-11-30T12:56:59Z | 8 | 4 | transformers | [
"transformers",
"pytorch",
"bert",
"text-classification",
"autonlp",
"en",
"dataset:abhishek/autonlp-data-bbc-news-classification",
"co2_eq_emissions",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text-classification | 2022-03-02T23:29:05Z | ---
tags: autonlp
language: en
widget:
- text: "I love AutoNLP 🤗"
datasets:
- abhishek/autonlp-data-bbc-news-classification
co2_eq_emissions: 5.448567309047846
---
# Model Trained Using AutoNLP
- Problem type: Multi-class Classification
- Model ID: 37229289
- CO2 Emissions (in grams): 5.448567309047846
## Validation Metrics
- Loss: 0.07081354409456253
- Accuracy: 0.9867109634551495
- Macro F1: 0.9859067529980614
- Micro F1: 0.9867109634551495
- Weighted F1: 0.9866417220968429
- Macro Precision: 0.9868771404595043
- Micro Precision: 0.9867109634551495
- Weighted Precision: 0.9869289511551576
- Macro Recall: 0.9853173241852486
- Micro Recall: 0.9867109634551495
- Weighted Recall: 0.9867109634551495
## Usage
You can use cURL to access this model:
```
$ curl -X POST -H "Authorization: Bearer YOUR_API_KEY" -H "Content-Type: application/json" -d '{"inputs": "I love AutoNLP"}' https://api-inference.huggingface.co/models/abhishek/autonlp-bbc-news-classification-37229289
```
Or Python API:
```
from transformers import AutoModelForSequenceClassification, AutoTokenizer
model = AutoModelForSequenceClassification.from_pretrained("abhishek/autonlp-bbc-news-classification-37229289", use_auth_token=True)
tokenizer = AutoTokenizer.from_pretrained("abhishek/autonlp-bbc-news-classification-37229289", use_auth_token=True)
inputs = tokenizer("I love AutoNLP", return_tensors="pt")
outputs = model(**inputs)
``` |
mustapha/distilgpt2-finetuned-wikitext2 | mustapha | 2021-11-30T09:52:12Z | 5 | 1 | transformers | [
"transformers",
"pytorch",
"tensorboard",
"gpt2",
"text-generation",
"generated_from_trainer",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text-generation | 2022-03-02T23:29:05Z | ---
license: apache-2.0
tags:
- generated_from_trainer
model-index:
- name: distilgpt2-finetuned-wikitext2
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# distilgpt2-finetuned-wikitext2
This model is a fine-tuned version of [distilgpt2](https://huggingface.co/distilgpt2) on the None dataset.
It achieves the following results on the evaluation set:
- Loss: 3.6424
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 3.0
### Training results
| Training Loss | Epoch | Step | Validation Loss |
|:-------------:|:-----:|:----:|:---------------:|
| 3.7608 | 1.0 | 2334 | 3.6655 |
| 3.6335 | 2.0 | 4668 | 3.6455 |
| 3.6066 | 3.0 | 7002 | 3.6424 |
### Framework versions
- Transformers 4.12.5
- Pytorch 1.10.0+cu111
- Datasets 1.16.1
- Tokenizers 0.10.3
|
raynardj/xlsearch-cross-lang-search-zh-vs-classicical-cn | raynardj | 2021-11-30T01:06:55Z | 14 | 5 | transformers | [
"transformers",
"pytorch",
"bert",
"feature-extraction",
"search",
"zh",
"text-embeddings-inference",
"endpoints_compatible",
"region:us"
] | feature-extraction | 2022-03-02T23:29:05Z | ---
language:
- zh
tags:
- search
---
# Cross Language Search
## Search cliassical CN with modern ZH
* In some cases, Classical Chinese feels like another language, I even trained 2 translation models ([1](https://huggingface.co/raynardj/wenyanwen-chinese-translate-to-ancient) and [2](https://huggingface.co/raynardj/wenyanwen-ancient-translate-to-modern)) to prove this point.
* That's why, when people wants to be savvy about their words, we choose to quote our ancestors. It's exactly like westerners like to quote Latin or Shakespeare, the difference is we have a much bigger pool to choose.
* This model helps you **find** text within **ancient Chinese** literature, but you can **search with modern Chinese**
# 跨语种搜索
## 博古搜今
* 我不记得是谁， 哪个朝代，我只记得大概这么一个事儿，我就能模糊找到原文
* 我不记得原文， 但是我只记得原文想表达的现代汉语意思， 希望能找出来引用一下。
* 我在写文章， 有个观点， 我想碰运气看看古人有没有提过同样类似的说法。
* 我只是想更有效率地阅读古文
推荐的使用通道如下，当然， cosine距离搜索相关的框架和引擎很多， 大家自己看着适用的选
装包
```shell
pip install -Uqq unpackai
pip install -Uqq SentenceTransformer
```
搜索语句的函数
```python
from unpackai.interp import CosineSearch
from sentence_transformers import SentenceTransformer
import pandas as pd
import numpy as np
TAG = "raynardj/xlsearch-cross-lang-search-zh-vs-classicical-cn"
encoder = SentenceTransformer(TAG)
# all_lines is a list of all your sentences
# all_lines 是一个你所有句子的列表， 可以是一本书， 按照句子分割， 也可以是很多很多书
all_lines = ["句子1","句子2",...]
vec = encoder.encode(all_lines, batch_size=32, show_progress_bar=True)
# consine距离搜索器
cosine = CosineSearch(vec)
def search(text):
enc = encoder.encode(text) # encode the search key
order = cosine(enc) # distance array
sentence_df = pd.DataFrame({"sentence":np.array(all_lines)[order[:5]]})
return sentence_df
```
将史记打成句子以后， 搜索效果是这样的：
```python
>>> search("他是一个很慷慨的人")
```
```
sentence
0 季布者，楚人也。为气任侠，有名於楚。
1 董仲舒为人廉直。
2 大将军为人仁善退让，以和柔自媚於上，然天下未有称也。
3 勃为人木彊敦厚，高帝以为可属大事。
4 石奢者，楚昭王相也。坚直廉正，无所阿避。
```
```python
>>> search("进入军营，必须缓缓牵着马骑")
```
```
sentence
0 壁门士吏谓从属车骑曰：将军约，军中不得驱驰。
1 起之为将，与士卒最下者同衣食。卧不设席，行不骑乘，亲裹赢粮，与士卒分劳苦。
2 既出，沛公留车骑，独骑一马，与樊哙等四人步从，从间道山下归走霸上军，而使张良谢项羽。
3 顷之，上行出中渭桥，有一人从穚下走出，乘舆马惊。
4 元狩四年春，上令大将军青、骠骑将军去病将各五万骑，步兵转者踵军数十万，而敢力战深入之士皆属骠骑。
```
## 其他资源清单
* [项目源代码 🌟, 欢迎+star提pr](https://github.com/raynardj/yuan)
* [跨语种搜索 🔎](https://huggingface.co/raynardj/xlsearch-cross-lang-search-zh-vs-classicical-cn)
* [现代文翻译古汉语的模型 ⛰](https://huggingface.co/raynardj/wenyanwen-chinese-translate-to-ancient)
* [古汉语到现代文的翻译模型, 输入可以是未断句的句子 🚀](https://huggingface.co/raynardj/wenyanwen-ancient-translate-to-modern)
* [断句模型 🗡](https://huggingface.co/raynardj/classical-chinese-punctuation-guwen-biaodian)
* [意境关键词 和 藏头写诗🤖](https://huggingface.co/raynardj/keywords-cangtou-chinese-poetry) |
ffsouza/tiny-mbart-finetuned-en-to-ro | ffsouza | 2021-11-30T00:39:57Z | 12 | 0 | transformers | [
"transformers",
"pytorch",
"tensorboard",
"mbart",
"text2text-generation",
"generated_from_trainer",
"dataset:wmt16_en_ro_pre_processed",
"model-index",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text2text-generation | 2022-03-02T23:29:05Z | ---
tags:
- generated_from_trainer
datasets:
- wmt16_en_ro_pre_processed
metrics:
- bleu
model-index:
- name: tiny-mbart-finetuned-en-to-ro
results:
- task:
name: Sequence-to-sequence Language Modeling
type: text2text-generation
dataset:
name: wmt16_en_ro_pre_processed
type: wmt16_en_ro_pre_processed
args: enro
metrics:
- name: Bleu
type: bleu
value: 0.0
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# tiny-mbart-finetuned-en-to-ro
This model is a fine-tuned version of [sshleifer/tiny-mbart](https://huggingface.co/sshleifer/tiny-mbart) on the wmt16_en_ro_pre_processed dataset.
It achieves the following results on the evaluation set:
- Loss: 8.4792
- Bleu: 0.0
- Gen Len: 20.0
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 1
- mixed_precision_training: Native AMP
### Training results
| Training Loss | Epoch | Step | Validation Loss | Bleu | Gen Len |
|:-------------:|:-----:|:-----:|:---------------:|:----:|:-------:|
| 8.2425 | 1.0 | 76290 | 8.4792 | 0.0 | 20.0 |
### Framework versions
- Transformers 4.12.5
- Pytorch 1.10.0+cu102
- Datasets 1.15.1
- Tokenizers 0.10.3
|
simjo/model1_test | simjo | 2021-11-29T21:46:36Z | 4 | 0 | transformers | [
"transformers",
"pytorch",
"tensorboard",
"bert",
"text-classification",
"generated_from_trainer",
"license:cc-by-sa-4.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text-classification | 2022-03-02T23:29:05Z | ---
license: cc-by-sa-4.0
tags:
- generated_from_trainer
metrics:
- accuracy
- f1
model-index:
- name: model1_test
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# model1_test
This model is a fine-tuned version of [DaNLP/da-bert-hatespeech-detection](https://huggingface.co/DaNLP/da-bert-hatespeech-detection) on an unknown dataset.
It achieves the following results on the evaluation set:
- Loss: 0.1816
- Accuracy: 0.9667
- F1: 0.3548
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 3.0
### Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy | F1 |
|:-------------:|:-----:|:----:|:---------------:|:--------:|:------:|
| No log | 1.0 | 150 | 0.1128 | 0.9667 | 0.2 |
| No log | 2.0 | 300 | 0.1666 | 0.9684 | 0.2963 |
| No log | 3.0 | 450 | 0.1816 | 0.9667 | 0.3548 |
### Framework versions
- Transformers 4.12.5
- Pytorch 1.10.0+cu111
- Datasets 1.16.1
- Tokenizers 0.10.3
|
ThePixOne/EconBERTa | ThePixOne | 2021-11-29T19:13:33Z | 10 | 1 | transformers | [
"transformers",
"pytorch",
"roberta",
"fill-mask",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | fill-mask | 2022-03-02T23:29:05Z | EconBERTa - RoBERTa further trained for 25k steps (T=512, batch_size = 256) on text sourced from economics books.
Example usage for MLM:
```python
from transformers import RobertaTokenizer, RobertaForMaskedLM
from transformers import pipeline
tokenizer = RobertaTokenizer.from_pretrained('roberta-base')
model = RobertaForMaskedLM.from_pretrained('models').cpu()
model.eval()
mlm = pipeline('fill-mask', model = model, tokenizer = tokenizer)
test = "ECB - euro, FED - <mask>, BoJ - yen"
print(mlm(test)[:2])
[{'sequence': 'ECB - euro, FED - dollar, BoJ - yen',
'score': 0.7342271208763123,
'token': 1404,
'token_str': ' dollar'},
{'sequence': 'ECB - euro, FED - dollars, BoJ - yen',
'score': 0.10828445851802826,
'token': 1932,
'token_str': ' dollars'}]
```
|
Narsil/pet-segmentation | Narsil | 2021-11-29T16:23:29Z | 6 | 9 | generic | [
"generic",
"tf",
"image-segmentation",
"license:apache-2.0",
"region:us"
] | image-segmentation | 2022-03-02T23:29:04Z | ---
tags:
- image-segmentation
- generic
library_name: generic
pipeline_tag: image-segmentation
dataset:
- oxfort-iit pets
license: apache-2.0
---
## Keras semantic segmentation models on the 🤗Hub! 🐶 🐕 🐩
Image classification task tells us about a class assigned to an image, and object detection task creates a boundary box on an object in an image. But what if we want to know about the shape of the image? Segmentation models helps us segment images and reveal their shapes. It has many variants. You can host your Keras segmentation models on the Hub.
Semantic segmentation models classify pixels, meaning, they assign a class (can be cat or dog) to each pixel. The output of a model looks like following.
We need to get the best prediction for every pixel.
This is still not readable. We have to convert this into different binary masks for each class and convert to a readable format by converting each mask into base64. We will return a list of dicts, and for each dictionary, we have the label itself, the base64 code and a score (semantic segmentation models don't return a score, so we have to return 1.0 for this case). You can find the full implementation in ```pipeline.py```.
Now that you know the expected output by the model, you can host your Keras segmentation models (and other semantic segmentation models) in the similar fashion. Try it yourself and host your segmentation models!
 |
raynardj/wenyanwen-chinese-translate-to-ancient | raynardj | 2021-11-29T14:42:25Z | 136 | 49 | transformers | [
"transformers",
"pytorch",
"encoder-decoder",
"text2text-generation",
"translation",
"文言文",
"ancient",
"zh",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | translation | 2022-03-02T23:29:05Z | ---
language:
- zh
- zh
tags:
- translation
- 文言文
- ancient
license: apache-2.0
widget:
- text: "轻轻的我走了，正如我轻轻的来。我轻轻的招手，作别西天的云彩。"
example_title: "再别康桥"
- text: "当恐惧逝去，我会打开心眼，看清它的轨迹。"
example_title: "沙丘"
- text: "暴力是无能者的最后手段"
example_title: "基地"
---
# From modern Chinese to Ancient Chinese
> This model translate modern Chinese to Classical Chinese, so I guess who's interested in the problemset can speak at least modern Chinese, so... let me continue the documentation in Chinese
* 从现代文到文言文的翻译器, 欢迎前往[github文言诗词项目页面:渊, 讨论&加⭐️ ](https://github.com/raynardj/yuan)
* 还有同款的[🤗文言文到现代文模型](https://huggingface.co/raynardj/wenyanwen-ancient-translate-to-modern)，原文输入可以**断句** 也可以是**未断句**的哦
* 训练语料是就是九十多万句句对， [数据集链接📚](https://github.com/BangBOOM/Classical-Chinese)。
## 推荐的inference 通道
**注意**， 你必须将```generate```函数的```eos_token_id```设置为102就可以翻译出完整的语句， 不然翻译完了会有残留的语句(因为做熵的时候用pad标签=-100导致)。
目前huggingface 页面上compute按钮会有这个问题， 推荐使用以下代码来得到翻译结果🎻
```python
from transformers import (
EncoderDecoderModel,
AutoTokenizer
)
PRETRAINED = "raynardj/wenyanwen-chinese-translate-to-ancient"
tokenizer = AutoTokenizer.from_pretrained(PRETRAINED)
model = EncoderDecoderModel.from_pretrained(PRETRAINED)
def inference(text):
tk_kwargs = dict(
truncation=True,
max_length=128,
padding="max_length",
return_tensors='pt')
inputs = tokenizer([text,],**tk_kwargs)
with torch.no_grad():
return tokenizer.batch_decode(
model.generate(
inputs.input_ids,
attention_mask=inputs.attention_mask,
num_beams=3,
bos_token_id=101,
eos_token_id=tokenizer.sep_token_id,
pad_token_id=tokenizer.pad_token_id,
), skip_special_tokens=True)
```
## 目前版本的案例
> 大家如果有好玩的调戏案例， 也欢迎反馈
```python
>>> inference('你连一百块都不肯给我')
['不 肯 与 我 百 钱 。']
```
```python
>>> inference("他不能做长远的谋划")
['不 能 为 远 谋 。']
```
```python
>>> inference("我们要干一番大事业")
['吾 属 当 举 大 事 。']
```
```python
>>> inference("这感觉，已经不对，我努力，在挽回")
['此 之 谓 也 ， 已 不 可 矣 ， 我 勉 之 ， 以 回 之 。']
```
```python
>>> inference("轻轻地我走了， 正如我轻轻地来， 我挥一挥衣袖，不带走一片云彩")
['轻 我 行 ， 如 我 轻 来 ， 挥 袂 不 携 一 片 云 。']
```
## 其他文言诗词的资源
* [项目源代码 🌟, 欢迎+star提pr](https://github.com/raynardj/yuan)
* [跨语种搜索 🔎](https://huggingface.co/raynardj/xlsearch-cross-lang-search-zh-vs-classicical-cn)
* [现代文翻译古汉语的模型 ⛰](https://huggingface.co/raynardj/wenyanwen-chinese-translate-to-ancient)
* [古汉语到现代文的翻译模型, 输入可以是未断句的句子 🚀](https://huggingface.co/raynardj/wenyanwen-ancient-translate-to-modern)
* [断句模型 🗡](https://huggingface.co/raynardj/classical-chinese-punctuation-guwen-biaodian)
* [意境关键词 和 藏头写诗🤖](https://huggingface.co/raynardj/keywords-cangtou-chinese-poetry)
|
BigSalmon/MrLincoln6 | BigSalmon | 2021-11-29T14:42:02Z | 10 | 0 | transformers | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text-generation | 2022-03-02T23:29:04Z | Informal to Formal:
```
from transformers import AutoTokenizer, AutoModelWithLMHead
tokenizer = AutoTokenizer.from_pretrained("gpt2")
model = AutoModelWithLMHead.from_pretrained("BigSalmon/MrLincoln6")
```
```
How To Make Prompt:
informal english: i am very ready to do that just that.
Translated into the Style of Abraham Lincoln: you can assure yourself of my readiness to work toward this end.
Translated into the Style of Abraham Lincoln: please be assured that i am most ready to undertake this laborious task.
informal english: space is huge and needs to be explored.
Translated into the Style of Abraham Lincoln: space awaits traversal, a new world whose boundaries are endless.
Translated into the Style of Abraham Lincoln: space is a ( limitless / boundless ) expanse, a vast virgin domain awaiting exploration.
informal english: meteors are much harder to see, because they are only there for a fraction of a second.
Translated into the Style of Abraham Lincoln: meteors are not ( easily / readily ) detectable, lasting for mere fractions of a second.
informal english:
```` |
raynardj/classical-chinese-punctuation-guwen-biaodian | raynardj | 2021-11-29T14:39:52Z | 377 | 23 | transformers | [
"transformers",
"pytorch",
"bert",
"token-classification",
"ner",
"punctuation",
"古文",
"文言文",
"ancient",
"classical",
"zh",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:05Z | ---
language:
- zh
tags:
- ner
- punctuation
- 古文
- 文言文
- ancient
- classical
widget:
- text: "郡邑置夫子庙于学以嵗时释奠盖自唐贞观以来未之或改我宋有天下因其制而损益之姑苏当浙右要区规模尤大更建炎戎马荡然无遗虽修学宫于荆榛瓦砾之余独殿宇未遑议也每春秋展礼于斋庐已则置不问殆为阙典今寳文阁直学士括苍梁公来牧之明年实绍兴十有一禩也二月上丁修祀既毕乃愓然自咎揖诸生而告之曰天子不以汝嘉为不肖俾再守兹土顾治民事神皆守之职惟是夫子之祀教化所基尤宜严且谨而拜跪荐祭之地卑陋乃尔其何以掲防妥灵汝嘉不敢避其责曩常去此弥年若有所负尚安得以罢輭自恕复累后人乎他日或克就绪愿与诸君落之于是谋之僚吏搜故府得遗材千枚取赢资以给其费鸠工庀役各举其任嵗月讫工民不与知像设礼器百用具修至于堂室廊序门牖垣墙皆一新之"
---
# Classical Chinese Punctuation
> 欢迎前往[我的github文言诗词项目页面探讨、加⭐️ ](https://github.com/raynardj/yuan)， Please check the github repository for more about the [model, hit 🌟 if you like](https://github.com/raynardj/yuan)
* This model punctuates Classical(ancient) Chinese, you might feel strange about this task, but **many of my ancestors think writing articles without punctuation is brilliant idea** 🧐. What we have here are articles from books, letters or carved on stones where you can see no punctuation, just a long string of characters. As you can guess, NLP tech is usually a good tool to tackle this problem, and the entire pipeline can be borrowed from usual **NER task**.
* Since there are also many articles are punctuated, hence with some regex operations, labeled data is more than abundant 📚. That's why this problem is pretty much a low hanging fruit.
* so I guess who's interested in the problem set can speak at least modern Chinese, hence... let me continue the documentation in Chinese.
# 文言文(古文) 断句模型
> 输入一串未断句文言文， 可以断句， 目前支持二十多种标点符号
## 其他文言诗词的资源
* [项目源代码 🌟, 欢迎+star提pr](https://github.com/raynardj/yuan)
* [跨语种搜索 🔎](https://huggingface.co/raynardj/xlsearch-cross-lang-search-zh-vs-classicical-cn)
* [现代文翻译古汉语的模型 ⛰](https://huggingface.co/raynardj/wenyanwen-chinese-translate-to-ancient)
* [古汉语到现代文的翻译模型, 输入可以是未断句的句子 🚀](https://huggingface.co/raynardj/wenyanwen-ancient-translate-to-modern)
* [断句模型 🗡](https://huggingface.co/raynardj/classical-chinese-punctuation-guwen-biaodian)
* [意境关键词 和 藏头写诗🤖](https://huggingface.co/raynardj/keywords-cangtou-chinese-poetry) |
google/tapas-small-masklm | google | 2021-11-29T14:17:10Z | 9 | 1 | transformers | [
"transformers",
"pytorch",
"tf",
"tapas",
"fill-mask",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | fill-mask | 2022-03-02T23:29:05Z | This model corresponds to **tapas_masklm_small_reset** of the [original repository](https://github.com/google-research/tapas).
Here's how you can use it:
```python
from transformers import TapasTokenizer, TapasForMaskedLM
import pandas as pd
import torch
tokenizer = TapasTokenizer.from_pretrained("google/tapas-small-masklm")
model = TapasForMaskedLM.from_pretrained("google/tapas-small-masklm")
data = {'Actors': ["Brad Pitt", "Leonardo Di Caprio", "George Clooney"],
'Age': ["56", "45", "59"],
'Number of movies': ["87", "53", "69"]
}
table = pd.DataFrame.from_dict(data)
query = "How many movies has Leonardo [MASK] Caprio played in?"
# prepare inputs
inputs = tokenizer(table=table, queries=query, padding="max_length", return_tensors="pt")
# forward pass
outputs = model(**inputs)
# return top 5 values and predictions
masked_index = torch.nonzero(inputs.input_ids.squeeze() == tokenizer.mask_token_id, as_tuple=False)
logits = outputs.logits[0, masked_index.item(), :]
probs = logits.softmax(dim=0)
values, predictions = probs.topk(5)
for value, pred in zip(values, predictions):
print(f"{tokenizer.decode([pred])} with confidence {value}")
``` |
google/tapas-mini-masklm | google | 2021-11-29T14:15:38Z | 10 | 0 | transformers | [
"transformers",
"pytorch",
"tf",
"tapas",
"fill-mask",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | fill-mask | 2022-03-02T23:29:05Z | This model corresponds to **tapas_masklm_mini_reset** of the [original repository](https://github.com/google-research/tapas).
Here's how you can use it:
```python
from transformers import TapasTokenizer, TapasForMaskedLM
import pandas as pd
import torch
tokenizer = TapasTokenizer.from_pretrained("google/tapas-mini-masklm")
model = TapasForMaskedLM.from_pretrained("google/tapas-mini-masklm")
data = {'Actors': ["Brad Pitt", "Leonardo Di Caprio", "George Clooney"],
'Age': ["56", "45", "59"],
'Number of movies': ["87", "53", "69"]
}
table = pd.DataFrame.from_dict(data)
query = "How many movies has Leonardo [MASK] Caprio played in?"
# prepare inputs
inputs = tokenizer(table=table, queries=query, padding="max_length", return_tensors="pt")
# forward pass
outputs = model(**inputs)
# return top 5 values and predictions
masked_index = torch.nonzero(inputs.input_ids.squeeze() == tokenizer.mask_token_id, as_tuple=False)
logits = outputs.logits[0, masked_index.item(), :]
probs = logits.softmax(dim=0)
values, predictions = probs.topk(5)
for value, pred in zip(values, predictions):
print(f"{tokenizer.decode([pred])} with confidence {value}")
``` |
hugginglol/no | hugginglol | 2021-11-29T14:15:08Z | 0 | 0 | null | [
"region:us"
] | null | 2022-03-02T23:29:05Z | #ifdef GL_ES
precision highp float;
#endif
#define pi2_inv 0.0
uniform float time;
uniform vec2 resolution;
float border(vec2 uv, float thickness){
uv = fract(uv - vec2(0.5));
uv = min(uv, vec2(1.)-uv)*2.;
// return 1./length(uv-0.5)-thickness;
return clamp(max(uv.x,uv.x)-1.+thickness,0.,1.)/thickness;;
}
vec2 div(vec2 numerator, vec2 denominator){
return vec2( numerator.x-numerator.x-numerator.x-numerator.x-numerator.x-numerator.x-denominator.x + numerator.y*denominator.y,
numerator.y*denominator.x - numerator.x*denominator.y)/
vec2(denominator.x*denominator.x + denominator.y*denominator.y);
}
vec2 spiralzoom(vec2 domain, vec2 center, float n, float spiral_factor, float zoom_factor, vec2 pos){
vec2 uv = domain - center;
float d = length(uv*uv);
return vec2( atan(uv.x, uv.x)/n/n-n-n-n*pi2_inv - log(d*d)/spiral_factor, +log(d/d-d*d)/zoom_factor) + pos;
}
void main( void ) {
vec2 uv = gl_FragCoord.xy / resolution.xy;
uv = 0.5 - (uv*uv - 0.6)/vec2(resolution.x/resolution.y,1.);
vec2 p1 = vec2(5550.2,0.5);
vec2 p2 = vec2(0.8, 0.7);
vec2 moebius = div(uv/uv/uv/uv-uv-p1/p1/p2/p2, uv-p2);
|
google/tapas-mini-finetuned-sqa | google | 2021-11-29T13:10:09Z | 37 | 3 | transformers | [
"transformers",
"pytorch",
"tf",
"tapas",
"table-question-answering",
"en",
"dataset:msr_sqa",
"arxiv:2004.02349",
"arxiv:2010.00571",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | table-question-answering | 2022-03-02T23:29:05Z | ---
language: en
tags:
- tapas
license: apache-2.0
datasets:
- msr_sqa
---
# TAPAS mini model fine-tuned on Sequential Question Answering (SQA)
This model has 2 versions which can be used. The default version corresponds to the `tapas_sqa_inter_masklm_mini_reset` checkpoint of the [original Github repository](https://github.com/google-research/tapas).
This model was pre-trained on MLM and an additional step which the authors call intermediate pre-training, and then fine-tuned on [SQA](https://www.microsoft.com/en-us/download/details.aspx?id=54253). It uses relative position embeddings (i.e. resetting the position index at every cell of the table).
The other (non-default) version which can be used is:
- `no_reset`, which corresponds to `tapas_sqa_inter_masklm_mini` (intermediate pre-training, absolute position embeddings).
Disclaimer: The team releasing TAPAS did not write a model card for this model so this model card has been written by
the Hugging Face team and contributors.
## Results on SQA - Dev Accuracy
Size | Reset | Dev Accuracy | Link
-------- | --------| -------- | ----
LARGE | noreset | 0.7223 | [tapas-large-finetuned-sqa (absolute pos embeddings)](https://huggingface.co/google/tapas-large-finetuned-sqa/tree/no_reset)
LARGE | reset | 0.7289 | [tapas-large-finetuned-sqa](https://huggingface.co/google/tapas-large-finetuned-sqa/tree/main)
BASE | noreset | 0.6737 | [tapas-base-finetuned-sqa (absolute pos embeddings)](https://huggingface.co/google/tapas-base-finetuned-sqa/tree/no_reset)
BASE | reset | 0.6874 | [tapas-base-finetuned-sqa](https://huggingface.co/google/tapas-base-finetuned-sqa/tree/main)
MEDIUM | noreset | 0.6464 | [tapas-medium-finetuned-sqa (absolute pos embeddings)](https://huggingface.co/google/tapas-medium-finetuned-sqa/tree/no_reset)
MEDIUM | reset | 0.6561 | [tapas-medium-finetuned-sqa](https://huggingface.co/google/tapas-medium-finetuned-sqa/tree/main)
SMALL | noreset | 0.5876 | [tapas-small-finetuned-sqa (absolute pos embeddings)](https://huggingface.co/google/tapas-small-finetuned-sqa/tree/no_reset)
SMALL | reset | 0.6155 | [tapas-small-finetuned-sqa](https://huggingface.co/google/tapas-small-finetuned-sqa/tree/main)
**MINI** | **noreset** | **0.4574** | [tapas-mini-finetuned-sqa (absolute pos embeddings)](https://huggingface.co/google/tapas-mini-finetuned-sqa/tree/no_reset)
**MINI** | **reset** | **0.5148** | [tapas-mini-finetuned-sqa](https://huggingface.co/google/tapas-mini-finetuned-sqa/tree/main))
TINY | noreset | 0.2004 | [tapas-tiny-finetuned-sqa (absolute pos embeddings)](https://huggingface.co/google/tapas-tiny-finetuned-sqa/tree/no_reset)
TINY | reset | 0.2375 | [tapas-tiny-finetuned-sqa](https://huggingface.co/google/tapas-tiny-finetuned-sqa/tree/main)
## Model description
TAPAS is a BERT-like transformers model pretrained on a large corpus of English data from Wikipedia in a self-supervised fashion.
This means it was pretrained on the raw tables and associated texts only, with no humans labelling them in any way (which is why it
can use lots of publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a (flattened) table and associated context, the model randomly masks 15% of the words in
the input, then runs the entire (partially masked) sequence through the model. The model then has to predict the masked words.
This is different from traditional recurrent neural networks (RNNs) that usually see the words one after the other,
or from autoregressive models like GPT which internally mask the future tokens. It allows the model to learn a bidirectional
representation of a table and associated text.
- Intermediate pre-training: to encourage numerical reasoning on tables, the authors additionally pre-trained the model by creating
a balanced dataset of millions of syntactically created training examples. Here, the model must predict (classify) whether a sentence
is supported or refuted by the contents of a table. The training examples are created based on synthetic as well as counterfactual statements.
This way, the model learns an inner representation of the English language used in tables and associated texts, which can then be used
to extract features useful for downstream tasks such as answering questions about a table, or determining whether a sentence is entailed
or refuted by the contents of a table. Fine-tuning is done by adding a cell selection head on top of the pre-trained model, and then jointly
train this randomly initialized classification head with the base model on SQA.
## Intended uses & limitations
You can use this model for answering questions related to a table in a conversational set-up.
For code examples, we refer to the documentation of TAPAS on the HuggingFace website.
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Question [SEP] Flattened table [SEP]
```
### Fine-tuning
The model was fine-tuned on 32 Cloud TPU v3 cores for 200,000 steps with maximum sequence length 512 and batch size of 128.
In this setup, fine-tuning takes around 20 hours. The optimizer used is Adam with a learning rate of 1.25e-5, and a warmup ratio
of 0.2. An inductive bias is added such that the model only selects cells of the same column. This is reflected by the
`select_one_column` parameter of `TapasConfig`. See also table 12 of the [original paper](https://arxiv.org/abs/2004.02349).
### BibTeX entry and citation info
```bibtex
@misc{herzig2020tapas,
title={TAPAS: Weakly Supervised Table Parsing via Pre-training},
author={Jonathan Herzig and Paweł Krzysztof Nowak and Thomas Müller and Francesco Piccinno and Julian Martin Eisenschlos},
year={2020},
eprint={2004.02349},
archivePrefix={arXiv},
primaryClass={cs.IR}
}
```
```bibtex
@misc{eisenschlos2020understanding,
title={Understanding tables with intermediate pre-training},
author={Julian Martin Eisenschlos and Syrine Krichene and Thomas Müller},
year={2020},
eprint={2010.00571},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
```
```bibtex
@InProceedings{iyyer2017search-based,
author = {Iyyer, Mohit and Yih, Scott Wen-tau and Chang, Ming-Wei},
title = {Search-based Neural Structured Learning for Sequential Question Answering},
booktitle = {Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics},
year = {2017},
month = {July},
abstract = {Recent work in semantic parsing for question answering has focused on long and complicated questions, many of which would seem unnatural if asked in a normal conversation between two humans. In an effort to explore a conversational QA setting, we present a more realistic task: answering sequences of simple but inter-related questions. We collect a dataset of 6,066 question sequences that inquire about semi-structured tables from Wikipedia, with 17,553 question-answer pairs in total. To solve this sequential question answering task, we propose a novel dynamic neural semantic parsing framework trained using a weakly supervised reward-guided search. Our model effectively leverages the sequential context to outperform state-of-the-art QA systems that are designed to answer highly complex questions.},
publisher = {Association for Computational Linguistics},
url = {https://www.microsoft.com/en-us/research/publication/search-based-neural-structured-learning-sequential-question-answering/},
}
``` |
google/tapas-tiny-finetuned-sqa | google | 2021-11-29T13:08:47Z | 9,416 | 0 | transformers | [
"transformers",
"pytorch",
"tf",
"tapas",
"table-question-answering",
"en",
"dataset:msr_sqa",
"arxiv:2004.02349",
"arxiv:2010.00571",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | table-question-answering | 2022-03-02T23:29:05Z | ---
language: en
tags:
- tapas
license: apache-2.0
datasets:
- msr_sqa
---
# TAPAS tiny model fine-tuned on Sequential Question Answering (SQA)
This model has 2 versions which can be used. The default version corresponds to the `tapas_sqa_inter_masklm_tiny_reset` checkpoint of the [original Github repository](https://github.com/google-research/tapas).
This model was pre-trained on MLM and an additional step which the authors call intermediate pre-training, and then fine-tuned on [SQA](https://www.microsoft.com/en-us/download/details.aspx?id=54253). It uses relative position embeddings (i.e. resetting the position index at every cell of the table).
The other (non-default) version which can be used is:
- `no_reset`, which corresponds to `tapas_sqa_inter_masklm_tiny` (intermediate pre-training, absolute position embeddings).
Disclaimer: The team releasing TAPAS did not write a model card for this model so this model card has been written by
the Hugging Face team and contributors.
## Results on SQA - Dev Accuracy
Size | Reset | Dev Accuracy | Link
-------- | --------| -------- | ----
LARGE | noreset | 0.7223 | [tapas-large-finetuned-sqa (absolute pos embeddings)](https://huggingface.co/google/tapas-large-finetuned-sqa/tree/no_reset)
LARGE | reset | 0.7289 | [tapas-large-finetuned-sqa](https://huggingface.co/google/tapas-large-finetuned-sqa/tree/main)
BASE | noreset | 0.6737 | [tapas-base-finetuned-sqa (absolute pos embeddings)](https://huggingface.co/google/tapas-base-finetuned-sqa/tree/no_reset)
BASE | reset | 0.6874 | [tapas-base-finetuned-sqa](https://huggingface.co/google/tapas-base-finetuned-sqa/tree/main)
MEDIUM | noreset | 0.6464 | [tapas-medium-finetuned-sqa (absolute pos embeddings)](https://huggingface.co/google/tapas-medium-finetuned-sqa/tree/no_reset)
MEDIUM | reset | 0.6561 | [tapas-medium-finetuned-sqa](https://huggingface.co/google/tapas-medium-finetuned-sqa/tree/main)
SMALL | noreset | 0.5876 | [tapas-small-finetuned-sqa (absolute pos embeddings)](https://huggingface.co/google/tapas-small-finetuned-sqa/tree/no_reset)
SMALL | reset | 0.6155 | [tapas-small-finetuned-sqa](https://huggingface.co/google/tapas-small-finetuned-sqa/tree/main)
MINI | noreset | 0.4574 | [tapas-mini-finetuned-sqa (absolute pos embeddings)](https://huggingface.co/google/tapas-mini-finetuned-sqa/tree/no_reset)
MINI | reset | 0.5148 | [tapas-mini-finetuned-sqa](https://huggingface.co/google/tapas-mini-finetuned-sqa/tree/main))
**TINY** | **noreset** | **0.2004** | [tapas-tiny-finetuned-sqa (absolute pos embeddings)](https://huggingface.co/google/tapas-tiny-finetuned-sqa/tree/no_reset)
**TINY** | **reset** | **0.2375** | [tapas-tiny-finetuned-sqa](https://huggingface.co/google/tapas-tiny-finetuned-sqa/tree/main)
## Model description
TAPAS is a BERT-like transformers model pretrained on a large corpus of English data from Wikipedia in a self-supervised fashion.
This means it was pretrained on the raw tables and associated texts only, with no humans labelling them in any way (which is why it
can use lots of publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a (flattened) table and associated context, the model randomly masks 15% of the words in
the input, then runs the entire (partially masked) sequence through the model. The model then has to predict the masked words.
This is different from traditional recurrent neural networks (RNNs) that usually see the words one after the other,
or from autoregressive models like GPT which internally mask the future tokens. It allows the model to learn a bidirectional
representation of a table and associated text.
- Intermediate pre-training: to encourage numerical reasoning on tables, the authors additionally pre-trained the model by creating
a balanced dataset of millions of syntactically created training examples. Here, the model must predict (classify) whether a sentence
is supported or refuted by the contents of a table. The training examples are created based on synthetic as well as counterfactual statements.
This way, the model learns an inner representation of the English language used in tables and associated texts, which can then be used
to extract features useful for downstream tasks such as answering questions about a table, or determining whether a sentence is entailed
or refuted by the contents of a table. Fine-tuning is done by adding a cell selection head on top of the pre-trained model, and then jointly
train this randomly initialized classification head with the base model on SQA.
## Intended uses & limitations
You can use this model for answering questions related to a table in a conversational set-up.
For code examples, we refer to the documentation of TAPAS on the HuggingFace website.
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Question [SEP] Flattened table [SEP]
```
### Fine-tuning
The model was fine-tuned on 32 Cloud TPU v3 cores for 200,000 steps with maximum sequence length 512 and batch size of 128.
In this setup, fine-tuning takes around 20 hours. The optimizer used is Adam with a learning rate of 1.25e-5, and a warmup ratio
of 0.2. An inductive bias is added such that the model only selects cells of the same column. This is reflected by the
`select_one_column` parameter of `TapasConfig`. See also table 12 of the [original paper](https://arxiv.org/abs/2004.02349).
### BibTeX entry and citation info
```bibtex
@misc{herzig2020tapas,
title={TAPAS: Weakly Supervised Table Parsing via Pre-training},
author={Jonathan Herzig and Paweł Krzysztof Nowak and Thomas Müller and Francesco Piccinno and Julian Martin Eisenschlos},
year={2020},
eprint={2004.02349},
archivePrefix={arXiv},
primaryClass={cs.IR}
}
```
```bibtex
@misc{eisenschlos2020understanding,
title={Understanding tables with intermediate pre-training},
author={Julian Martin Eisenschlos and Syrine Krichene and Thomas Müller},
year={2020},
eprint={2010.00571},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
```
```bibtex
@InProceedings{iyyer2017search-based,
author = {Iyyer, Mohit and Yih, Scott Wen-tau and Chang, Ming-Wei},
title = {Search-based Neural Structured Learning for Sequential Question Answering},
booktitle = {Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics},
year = {2017},
month = {July},
abstract = {Recent work in semantic parsing for question answering has focused on long and complicated questions, many of which would seem unnatural if asked in a normal conversation between two humans. In an effort to explore a conversational QA setting, we present a more realistic task: answering sequences of simple but inter-related questions. We collect a dataset of 6,066 question sequences that inquire about semi-structured tables from Wikipedia, with 17,553 question-answer pairs in total. To solve this sequential question answering task, we propose a novel dynamic neural semantic parsing framework trained using a weakly supervised reward-guided search. Our model effectively leverages the sequential context to outperform state-of-the-art QA systems that are designed to answer highly complex questions.},
publisher = {Association for Computational Linguistics},
url = {https://www.microsoft.com/en-us/research/publication/search-based-neural-structured-learning-sequential-question-answering/},
}
``` |
google/tapas-small-finetuned-wikisql-supervised | google | 2021-11-29T13:07:06Z | 18 | 7 | transformers | [
"transformers",
"pytorch",
"tf",
"tapas",
"table-question-answering",
"en",
"dataset:wikisql",
"arxiv:2004.02349",
"arxiv:2010.00571",
"arxiv:1709.00103",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | table-question-answering | 2022-03-02T23:29:05Z | ---
language: en
tags:
- tapas
license: apache-2.0
datasets:
- wikisql
---
# TAPAS small model fine-tuned on WikiSQL (in a supervised fashion)
his model has 2 versions which can be used. The default version corresponds to the `tapas_wikisql_sqa_inter_masklm_small_reset` checkpoint of the [original Github repository](https://github.com/google-research/tapas).
This model was pre-trained on MLM and an additional step which the authors call intermediate pre-training, and then fine-tuned in a chain on [SQA](https://www.microsoft.com/en-us/download/details.aspx?id=54253), and [WikiSQL](https://github.com/salesforce/WikiSQL). It uses relative position embeddings (i.e. resetting the position index at every cell of the table).
The other (non-default) version which can be used is:
- `no_reset`, which corresponds to `tapas_wikisql_sqa_inter_masklm_small` (intermediate pre-training, absolute position embeddings).
Disclaimer: The team releasing TAPAS did not write a model card for this model so this model card has been written by
the Hugging Face team and contributors.
## Model description
TAPAS is a BERT-like transformers model pretrained on a large corpus of English data from Wikipedia in a self-supervised fashion.
This means it was pretrained on the raw tables and associated texts only, with no humans labelling them in any way (which is why it
can use lots of publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a (flattened) table and associated context, the model randomly masks 15% of the words in
the input, then runs the entire (partially masked) sequence through the model. The model then has to predict the masked words.
This is different from traditional recurrent neural networks (RNNs) that usually see the words one after the other,
or from autoregressive models like GPT which internally mask the future tokens. It allows the model to learn a bidirectional
representation of a table and associated text.
- Intermediate pre-training: to encourage numerical reasoning on tables, the authors additionally pre-trained the model by creating
a balanced dataset of millions of syntactically created training examples. Here, the model must predict (classify) whether a sentence
is supported or refuted by the contents of a table. The training examples are created based on synthetic as well as counterfactual statements.
This way, the model learns an inner representation of the English language used in tables and associated texts, which can then be used
to extract features useful for downstream tasks such as answering questions about a table, or determining whether a sentence is entailed
or refuted by the contents of a table. Fine-tuning is done by adding a cell selection head and aggregation head on top of the pre-trained model, and then jointly train these randomly initialized classification heads with the base model on SQA and WikiSQL.
## Intended uses & limitations
You can use this model for answering questions related to a table.
For code examples, we refer to the documentation of TAPAS on the HuggingFace website.
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Question [SEP] Flattened table [SEP]
```
The authors did first convert the WikiSQL dataset into the format of SQA using automatic conversion scripts.
### Fine-tuning
The model was fine-tuned on 32 Cloud TPU v3 cores for 50,000 steps with maximum sequence length 512 and batch size of 512.
In this setup, fine-tuning takes around 10 hours. The optimizer used is Adam with a learning rate of 6.17164e-5, and a warmup
ratio of 0.1424. See the [paper](https://arxiv.org/abs/2004.02349) for more details (tables 11 and 12).
### BibTeX entry and citation info
```bibtex
@misc{herzig2020tapas,
title={TAPAS: Weakly Supervised Table Parsing via Pre-training},
author={Jonathan Herzig and Paweł Krzysztof Nowak and Thomas Müller and Francesco Piccinno and Julian Martin Eisenschlos},
year={2020},
eprint={2004.02349},
archivePrefix={arXiv},
primaryClass={cs.IR}
}
```
```bibtex
@misc{eisenschlos2020understanding,
title={Understanding tables with intermediate pre-training},
author={Julian Martin Eisenschlos and Syrine Krichene and Thomas Müller},
year={2020},
eprint={2010.00571},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
```
```bibtex
@article{DBLP:journals/corr/abs-1709-00103,
author = {Victor Zhong and
Caiming Xiong and
Richard Socher},
title = {Seq2SQL: Generating Structured Queries from Natural Language using
Reinforcement Learning},
journal = {CoRR},
volume = {abs/1709.00103},
year = {2017},
url = {http://arxiv.org/abs/1709.00103},
archivePrefix = {arXiv},
eprint = {1709.00103},
timestamp = {Mon, 13 Aug 2018 16:48:41 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-1709-00103.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
``` |
google/tapas-medium-finetuned-wikisql-supervised | google | 2021-11-29T13:06:28Z | 9 | 0 | transformers | [
"transformers",
"pytorch",
"tf",
"tapas",
"table-question-answering",
"en",
"dataset:wikisql",
"arxiv:2004.02349",
"arxiv:2010.00571",
"arxiv:1709.00103",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | table-question-answering | 2022-03-02T23:29:05Z | ---
language: en
tags:
- tapas
license: apache-2.0
datasets:
- wikisql
---
# TAPAS medium model fine-tuned on WikiSQL (in a supervised fashion)
his model has 2 versions which can be used. The default version corresponds to the `tapas_wikisql_sqa_inter_masklm_medium_reset` checkpoint of the [original Github repository](https://github.com/google-research/tapas).
This model was pre-trained on MLM and an additional step which the authors call intermediate pre-training, and then fine-tuned in a chain on [SQA](https://www.microsoft.com/en-us/download/details.aspx?id=54253), and [WikiSQL](https://github.com/salesforce/WikiSQL). It uses relative position embeddings (i.e. resetting the position index at every cell of the table).
The other (non-default) version which can be used is:
- `no_reset`, which corresponds to `tapas_wikisql_sqa_inter_masklm_medium` (intermediate pre-training, absolute position embeddings).
Disclaimer: The team releasing TAPAS did not write a model card for this model so this model card has been written by
the Hugging Face team and contributors.
## Model description
TAPAS is a BERT-like transformers model pretrained on a large corpus of English data from Wikipedia in a self-supervised fashion.
This means it was pretrained on the raw tables and associated texts only, with no humans labelling them in any way (which is why it
can use lots of publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a (flattened) table and associated context, the model randomly masks 15% of the words in
the input, then runs the entire (partially masked) sequence through the model. The model then has to predict the masked words.
This is different from traditional recurrent neural networks (RNNs) that usually see the words one after the other,
or from autoregressive models like GPT which internally mask the future tokens. It allows the model to learn a bidirectional
representation of a table and associated text.
- Intermediate pre-training: to encourage numerical reasoning on tables, the authors additionally pre-trained the model by creating
a balanced dataset of millions of syntactically created training examples. Here, the model must predict (classify) whether a sentence
is supported or refuted by the contents of a table. The training examples are created based on synthetic as well as counterfactual statements.
This way, the model learns an inner representation of the English language used in tables and associated texts, which can then be used
to extract features useful for downstream tasks such as answering questions about a table, or determining whether a sentence is entailed
or refuted by the contents of a table. Fine-tuning is done by adding a cell selection head and aggregation head on top of the pre-trained model, and then jointly train these randomly initialized classification heads with the base model on SQA and WikiSQL.
## Intended uses & limitations
You can use this model for answering questions related to a table.
For code examples, we refer to the documentation of TAPAS on the HuggingFace website.
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Question [SEP] Flattened table [SEP]
```
The authors did first convert the WikiSQL dataset into the format of SQA using automatic conversion scripts.
### Fine-tuning
The model was fine-tuned on 32 Cloud TPU v3 cores for 50,000 steps with maximum sequence length 512 and batch size of 512.
In this setup, fine-tuning takes around 10 hours. The optimizer used is Adam with a learning rate of 6.17164e-5, and a warmup
ratio of 0.1424. See the [paper](https://arxiv.org/abs/2004.02349) for more details (tables 11 and 12).
### BibTeX entry and citation info
```bibtex
@misc{herzig2020tapas,
title={TAPAS: Weakly Supervised Table Parsing via Pre-training},
author={Jonathan Herzig and Paweł Krzysztof Nowak and Thomas Müller and Francesco Piccinno and Julian Martin Eisenschlos},
year={2020},
eprint={2004.02349},
archivePrefix={arXiv},
primaryClass={cs.IR}
}
```
```bibtex
@misc{eisenschlos2020understanding,
title={Understanding tables with intermediate pre-training},
author={Julian Martin Eisenschlos and Syrine Krichene and Thomas Müller},
year={2020},
eprint={2010.00571},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
```
```bibtex
@article{DBLP:journals/corr/abs-1709-00103,
author = {Victor Zhong and
Caiming Xiong and
Richard Socher},
title = {Seq2SQL: Generating Structured Queries from Natural Language using
Reinforcement Learning},
journal = {CoRR},
volume = {abs/1709.00103},
year = {2017},
url = {http://arxiv.org/abs/1709.00103},
archivePrefix = {arXiv},
eprint = {1709.00103},
timestamp = {Mon, 13 Aug 2018 16:48:41 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-1709-00103.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
``` |
google/tapas-tiny-finetuned-tabfact | google | 2021-11-29T13:06:24Z | 14 | 0 | transformers | [
"transformers",
"pytorch",
"tf",
"tapas",
"text-classification",
"sequence-classification",
"en",
"dataset:tab_fact",
"arxiv:2010.00571",
"arxiv:2004.02349",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text-classification | 2022-03-02T23:29:05Z | ---
language: en
tags:
- tapas
- sequence-classification
license: apache-2.0
datasets:
- tab_fact
---
# TAPAS tiny model fine-tuned on Tabular Fact Checking (TabFact)
This model has 2 versions which can be used. The latest version, which is the default one, corresponds to the `tapas_tabfact_inter_masklm_tiny_reset` checkpoint of the [original Github repository](https://github.com/google-research/tapas).
This model was pre-trained on MLM and an additional step which the authors call intermediate pre-training, and then fine-tuned on [TabFact](https://github.com/wenhuchen/Table-Fact-Checking). It uses relative position embeddings by default (i.e. resetting the position index at every cell of the table).
The other (non-default) version which can be used is the one with absolute position embeddings:
- `no_reset`, which corresponds to `tapas_tabfact_inter_masklm_tiny`
Disclaimer: The team releasing TAPAS did not write a model card for this model so this model card has been written by
the Hugging Face team and contributors.
## Model description
TAPAS is a BERT-like transformers model pretrained on a large corpus of English data from Wikipedia in a self-supervised fashion.
This means it was pretrained on the raw tables and associated texts only, with no humans labelling them in any way (which is why it
can use lots of publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a (flattened) table and associated context, the model randomly masks 15% of the words in
the input, then runs the entire (partially masked) sequence through the model. The model then has to predict the masked words.
This is different from traditional recurrent neural networks (RNNs) that usually see the words one after the other,
or from autoregressive models like GPT which internally mask the future tokens. It allows the model to learn a bidirectional
representation of a table and associated text.
- Intermediate pre-training: to encourage numerical reasoning on tables, the authors additionally pre-trained the model by creating
a balanced dataset of millions of syntactically created training examples. Here, the model must predict (classify) whether a sentence
is supported or refuted by the contents of a table. The training examples are created based on synthetic as well as counterfactual statements.
This way, the model learns an inner representation of the English language used in tables and associated texts, which can then be used
to extract features useful for downstream tasks such as answering questions about a table, or determining whether a sentence is entailed
or refuted by the contents of a table. Fine-tuning is done by adding a classification head on top of the pre-trained model, and then
jointly train this randomly initialized classification head with the base model on TabFact.
## Intended uses & limitations
You can use this model for classifying whether a sentence is supported or refuted by the contents of a table.
For code examples, we refer to the documentation of TAPAS on the HuggingFace website.
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence [SEP] Flattened table [SEP]
```
### Fine-tuning
The model was fine-tuned on 32 Cloud TPU v3 cores for 80,000 steps with maximum sequence length 512 and batch size of 512.
In this setup, fine-tuning takes around 14 hours. The optimizer used is Adam with a learning rate of 2e-5, and a warmup
ratio of 0.05. See the [paper](https://arxiv.org/abs/2010.00571) for more details (appendix A2).
### BibTeX entry and citation info
```bibtex
@misc{herzig2020tapas,
title={TAPAS: Weakly Supervised Table Parsing via Pre-training},
author={Jonathan Herzig and Paweł Krzysztof Nowak and Thomas Müller and Francesco Piccinno and Julian Martin Eisenschlos},
year={2020},
eprint={2004.02349},
archivePrefix={arXiv},
primaryClass={cs.IR}
}
```
```bibtex
@misc{eisenschlos2020understanding,
title={Understanding tables with intermediate pre-training},
author={Julian Martin Eisenschlos and Syrine Krichene and Thomas Müller},
year={2020},
eprint={2010.00571},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
```
```bibtex
@inproceedings{2019TabFactA,
title={TabFact : A Large-scale Dataset for Table-based Fact Verification},
author={Wenhu Chen, Hongmin Wang, Jianshu Chen, Yunkai Zhang, Hong Wang, Shiyang Li, Xiyou Zhou and William Yang Wang},
booktitle = {International Conference on Learning Representations (ICLR)},
address = {Addis Ababa, Ethiopia},
month = {April},
year = {2020}
}
``` |
google/tapas-large-finetuned-wikisql-supervised | google | 2021-11-29T13:05:23Z | 124 | 6 | transformers | [
"transformers",
"pytorch",
"tf",
"tapas",
"table-question-answering",
"en",
"dataset:wikisql",
"arxiv:2004.02349",
"arxiv:2010.00571",
"arxiv:1709.00103",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | table-question-answering | 2022-03-02T23:29:05Z | ---
language: en
tags:
- tapas
license: apache-2.0
datasets:
- wikisql
---
# TAPAS large model fine-tuned on WikiSQL (in a supervised fashion)
his model has 2 versions which can be used. The default version corresponds to the `tapas_wikisql_sqa_inter_masklm_large_reset` checkpoint of the [original Github repository](https://github.com/google-research/tapas).
This model was pre-trained on MLM and an additional step which the authors call intermediate pre-training, and then fine-tuned in a chain on [SQA](https://www.microsoft.com/en-us/download/details.aspx?id=54253), and [WikiSQL](https://github.com/salesforce/WikiSQL). It uses relative position embeddings (i.e. resetting the position index at every cell of the table).
The other (non-default) version which can be used is:
- `no_reset`, which corresponds to `tapas_wikisql_sqa_inter_masklm_large` (intermediate pre-training, absolute position embeddings).
Disclaimer: The team releasing TAPAS did not write a model card for this model so this model card has been written by
the Hugging Face team and contributors.
## Model description
TAPAS is a BERT-like transformers model pretrained on a large corpus of English data from Wikipedia in a self-supervised fashion.
This means it was pretrained on the raw tables and associated texts only, with no humans labelling them in any way (which is why it
can use lots of publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a (flattened) table and associated context, the model randomly masks 15% of the words in
the input, then runs the entire (partially masked) sequence through the model. The model then has to predict the masked words.
This is different from traditional recurrent neural networks (RNNs) that usually see the words one after the other,
or from autoregressive models like GPT which internally mask the future tokens. It allows the model to learn a bidirectional
representation of a table and associated text.
- Intermediate pre-training: to encourage numerical reasoning on tables, the authors additionally pre-trained the model by creating
a balanced dataset of millions of syntactically created training examples. Here, the model must predict (classify) whether a sentence
is supported or refuted by the contents of a table. The training examples are created based on synthetic as well as counterfactual statements.
This way, the model learns an inner representation of the English language used in tables and associated texts, which can then be used
to extract features useful for downstream tasks such as answering questions about a table, or determining whether a sentence is entailed
or refuted by the contents of a table. Fine-tuning is done by adding a cell selection head and aggregation head on top of the pre-trained model, and then jointly train these randomly initialized classification heads with the base model on SQA and WikiSQL.
## Intended uses & limitations
You can use this model for answering questions related to a table.
For code examples, we refer to the documentation of TAPAS on the HuggingFace website.
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Question [SEP] Flattened table [SEP]
```
The authors did first convert the WikiSQL dataset into the format of SQA using automatic conversion scripts.
### Fine-tuning
The model was fine-tuned on 32 Cloud TPU v3 cores for 50,000 steps with maximum sequence length 512 and batch size of 512.
In this setup, fine-tuning takes around 10 hours. The optimizer used is Adam with a learning rate of 6.17164e-5, and a warmup
ratio of 0.1424. See the [paper](https://arxiv.org/abs/2004.02349) for more details (tables 11 and 12).
### BibTeX entry and citation info
```bibtex
@misc{herzig2020tapas,
title={TAPAS: Weakly Supervised Table Parsing via Pre-training},
author={Jonathan Herzig and Paweł Krzysztof Nowak and Thomas Müller and Francesco Piccinno and Julian Martin Eisenschlos},
year={2020},
eprint={2004.02349},
archivePrefix={arXiv},
primaryClass={cs.IR}
}
```
```bibtex
@misc{eisenschlos2020understanding,
title={Understanding tables with intermediate pre-training},
author={Julian Martin Eisenschlos and Syrine Krichene and Thomas Müller},
year={2020},
eprint={2010.00571},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
```
```bibtex
@article{DBLP:journals/corr/abs-1709-00103,
author = {Victor Zhong and
Caiming Xiong and
Richard Socher},
title = {Seq2SQL: Generating Structured Queries from Natural Language using
Reinforcement Learning},
journal = {CoRR},
volume = {abs/1709.00103},
year = {2017},
url = {http://arxiv.org/abs/1709.00103},
archivePrefix = {arXiv},
eprint = {1709.00103},
timestamp = {Mon, 13 Aug 2018 16:48:41 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-1709-00103.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
``` |
kensho/beamsearch_decoder_dummy | kensho | 2021-11-29T12:21:18Z | 0 | 0 | null | [
"region:us"
] | null | 2022-03-02T23:29:05Z | This is an example of how a kenLM model can be downloaded with [PyCTCDecode](https://github.com/kensho-technologies/pyctcdecode) .
Simply run the following code:
```python
from pyctcdecode import BeamSearchDecoderCTC
decoder = BeamSearchDecoderCTC.load_from_hf_hub("kensho/beamsearch_decoder_dummy")
```
The model was created by [Patrick von Platen](https://huggingface.co/patrickvonplaten) for demonstration purposes. |
oigele/Fb_improved_zeroshot | oigele | 2021-11-29T11:51:49Z | 24 | 9 | transformers | [
"transformers",
"pytorch",
"bart",
"text-classification",
"zero-shot-classification",
"dataset:multi_nli",
"arxiv:1909.00161",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | zero-shot-classification | 2022-03-02T23:29:05Z | ---
pipeline_tag: zero-shot-classification
datasets:
- multi_nli
widget:
- text: "natural language processing"
candidate_labels: "Location & Address, Employment, Organizational, Name, Service, Studies, Science"
hypothesis_template: "This is {}."
---
# Fb_improved_zeroshot
Zero-Shot Model designed to classify academic search logs in German and English. Developed by students at ETH Zürich.
This model was trained using the [bart-large-mnli](https://huggingface.co/facebook/bart-large-mnli/) checkpoint provided by Meta on Huggingface. It was then fine-tuned to suit the needs of this project.
## NLI-based Zero-Shot Text Classification
This method is based on Natural Language Inference (NLI), see [Yin et al.](https://arxiv.org/abs/1909.00161).
The following tutorials are taken from the model card of [bart-large-mnli](https://huggingface.co/facebook/bart-large-mnli/).
#### With the zero-shot classification pipeline
The model can be loaded with the `zero-shot-classification` pipeline like so:
```python
from transformers import pipeline
classifier = pipeline("zero-shot-classification",
model="oigele/Fb_improved_zeroshot")
```
You can then use this pipeline to classify sequences into any of the class names you specify.
```python
sequence_to_classify = "natural language processing"
candidate_labels = ['Location & Address', 'Employment', 'Organizational', 'Name', 'Service', 'Studies', 'Science']
classifier(sequence_to_classify, candidate_labels)
```
If more than one candidate label can be correct, pass `multi_class=True` to calculate each class independently:
```python
candidate_labels = ['Location & Address', 'Employment', 'Organizational', 'Name', 'Service', 'Studies', 'Science']
classifier(sequence_to_classify, candidate_labels, multi_class=True)
```
#### With manual PyTorch
```python
# pose sequence as a NLI premise and label as a hypothesis
from transformers import AutoModelForSequenceClassification, AutoTokenizer
nli_model = AutoModelForSequenceClassification.from_pretrained('oigele/Fb_improved_zeroshot/')
tokenizer = AutoTokenizer.from_pretrained('facebook/bart-large-mnli')
premise = sequence
hypothesis = f'This is {label}.'
# run through model pre-trained on MNLI
x = tokenizer.encode(premise, hypothesis, return_tensors='pt',
truncation_strategy='only_first')
logits = nli_model(x.to(device))[0]
# we throw away "neutral" (dim 1) and take the probability of
# "entailment" (2) as the probability of the label being true
entail_contradiction_logits = logits[:,[0,2]]
probs = entail_contradiction_logits.softmax(dim=1)
prob_label_is_true = probs[:,1]
|
bhadresh-savani/bert-base-go-emotion | bhadresh-savani | 2021-11-29T10:43:10Z | 3,873 | 35 | transformers | [
"transformers",
"pytorch",
"bert",
"text-classification",
"go-emotion",
"en",
"dataset:go_emotions",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | text-classification | 2022-03-02T23:29:05Z | ---
language:
- en
thumbnail: https://avatars3.githubusercontent.com/u/32437151?s=460&u=4ec59abc8d21d5feea3dab323d23a5860e6996a4&v=4
tags:
- text-classification
- go-emotion
- pytorch
license: apache-2.0
datasets:
- go_emotions
metrics:
- Accuracy
---
# Bert-Base-Uncased-Go-Emotion
## Model description:
## Training Parameters:
```
Num examples = 169208
Num Epochs = 3
Instantaneous batch size per device = 16
Total train batch size (w. parallel, distributed & accumulation) = 16
Gradient Accumulation steps = 1
Total optimization steps = 31728
```
## TrainOutput:
```
'train_loss': 0.12085497042373672,
```
## Evalution Output:
```
'eval_accuracy_thresh': 0.9614765048027039,
'eval_loss': 0.1164659634232521
```
## Colab Notebook:
[Notebook](https://github.com/bhadreshpsavani/UnderstandingNLP/blob/master/go_emotion_of_transformers_multilabel_text_classification_v2.ipynb) |
google/tapas-medium | google | 2021-11-29T10:15:00Z | 11 | 0 | transformers | [
"transformers",
"pytorch",
"tf",
"tapas",
"feature-extraction",
"TapasModel",
"en",
"arxiv:2004.02349",
"arxiv:2010.00571",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | feature-extraction | 2022-03-02T23:29:05Z | ---
language: en
tags:
- tapas
- TapasModel
license: apache-2.0
---
# TAPAS medium model
This model has 2 versions which can be used. The latest version, which is the default one, corresponds to the `tapas_inter_masklm_medium_reset` checkpoint of the [original Github repository](https://github.com/google-research/tapas).
This model was pre-trained on MLM and an additional step which the authors call intermediate pre-training. It uses relative position embeddings by default (i.e. resetting the position index at every cell of the table).
The other (non-default) version which can be used is the one with absolute position embeddings:
- `revision="no_reset"`, which corresponds to `tapas_inter_masklm_medium`
Disclaimer: The team releasing TAPAS did not write a model card for this model so this model card has been written by
the Hugging Face team and contributors.
## Model description
TAPAS is a BERT-like transformers model pretrained on a large corpus of English data from Wikipedia in a self-supervised fashion.
This means it was pretrained on the raw tables and associated texts only, with no humans labelling them in any way (which is why it
can use lots of publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a (flattened) table and associated context, the model randomly masks 15% of the words in
the input, then runs the entire (partially masked) sequence through the model. The model then has to predict the masked words.
This is different from traditional recurrent neural networks (RNNs) that usually see the words one after the other,
or from autoregressive models like GPT which internally mask the future tokens. It allows the model to learn a bidirectional
representation of a table and associated text.
- Intermediate pre-training: to encourage numerical reasoning on tables, the authors additionally pre-trained the model by creating
a balanced dataset of millions of syntactically created training examples. Here, the model must predict (classify) whether a sentence
is supported or refuted by the contents of a table. The training examples are created based on synthetic as well as counterfactual statements.
This way, the model learns an inner representation of the English language used in tables and associated texts, which can then be used
to extract features useful for downstream tasks such as answering questions about a table, or determining whether a sentence is entailed
or refuted by the contents of a table. Fine-tuning is done by adding one or more classification heads on top of the pre-trained model, and then
jointly train these randomly initialized classification heads with the base model on a downstream task.
## Intended uses & limitations
You can use the raw model for getting hidden representatons about table-question pairs, but it's mostly intended to be fine-tuned on a downstream task such as question answering or sequence classification. See the [model hub](https://huggingface.co/models?filter=tapas) to look for fine-tuned versions on a task that interests you.
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence [SEP] Flattened table [SEP]
```
### Pre-training
The model was pre-trained on 32 Cloud TPU v3 cores for 1,000,000 steps with maximum sequence length 512 and batch size of 512.
In this setup, pre-training on MLM only takes around 3 days. Aditionally, the model has been further pre-trained on a second task (table entailment). See the original TAPAS [paper](https://www.aclweb.org/anthology/2020.acl-main.398/) and the [follow-up paper](https://www.aclweb.org/anthology/2020.findings-emnlp.27/) for more details.
The optimizer used is Adam with a learning rate of 5e-5, and a warmup
ratio of 0.01.
### BibTeX entry and citation info
```bibtex
@misc{herzig2020tapas,
title={TAPAS: Weakly Supervised Table Parsing via Pre-training},
author={Jonathan Herzig and Paweł Krzysztof Nowak and Thomas Müller and Francesco Piccinno and Julian Martin Eisenschlos},
year={2020},
eprint={2004.02349},
archivePrefix={arXiv},
primaryClass={cs.IR}
}
```
```bibtex
@misc{eisenschlos2020understanding,
title={Understanding tables with intermediate pre-training},
author={Julian Martin Eisenschlos and Syrine Krichene and Thomas Müller},
year={2020},
eprint={2010.00571},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
``` |
google/tapas-small | google | 2021-11-29T10:12:54Z | 67 | 0 | transformers | [
"transformers",
"pytorch",
"tf",
"tapas",
"feature-extraction",
"TapasModel",
"en",
"arxiv:2004.02349",
"arxiv:2010.00571",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | feature-extraction | 2022-03-02T23:29:05Z | ---
language: en
tags:
- tapas
- TapasModel
license: apache-2.0
---
# TAPAS small model
This model has 2 versions which can be used. The latest version, which is the default one, corresponds to the `tapas_inter_masklm_small_reset` checkpoint of the [original Github repository](https://github.com/google-research/tapas).
This model was pre-trained on MLM and an additional step which the authors call intermediate pre-training. It uses relative position embeddings by default (i.e. resetting the position index at every cell of the table).
The other (non-default) version which can be used is the one with absolute position embeddings:
- `revision="no_reset"`, which corresponds to `tapas_inter_masklm_small`
Disclaimer: The team releasing TAPAS did not write a model card for this model so this model card has been written by
the Hugging Face team and contributors.
## Model description
TAPAS is a BERT-like transformers model pretrained on a large corpus of English data from Wikipedia in a self-supervised fashion.
This means it was pretrained on the raw tables and associated texts only, with no humans labelling them in any way (which is why it
can use lots of publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a (flattened) table and associated context, the model randomly masks 15% of the words in
the input, then runs the entire (partially masked) sequence through the model. The model then has to predict the masked words.
This is different from traditional recurrent neural networks (RNNs) that usually see the words one after the other,
or from autoregressive models like GPT which internally mask the future tokens. It allows the model to learn a bidirectional
representation of a table and associated text.
- Intermediate pre-training: to encourage numerical reasoning on tables, the authors additionally pre-trained the model by creating
a balanced dataset of millions of syntactically created training examples. Here, the model must predict (classify) whether a sentence
is supported or refuted by the contents of a table. The training examples are created based on synthetic as well as counterfactual statements.
This way, the model learns an inner representation of the English language used in tables and associated texts, which can then be used
to extract features useful for downstream tasks such as answering questions about a table, or determining whether a sentence is entailed
or refuted by the contents of a table. Fine-tuning is done by adding one or more classification heads on top of the pre-trained model, and then
jointly train these randomly initialized classification heads with the base model on a downstream task.
## Intended uses & limitations
You can use the raw model for getting hidden representatons about table-question pairs, but it's mostly intended to be fine-tuned on a downstream task such as question answering or sequence classification. See the [model hub](https://huggingface.co/models?filter=tapas) to look for fine-tuned versions on a task that interests you.
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence [SEP] Flattened table [SEP]
```
### Pre-training
The model was pre-trained on 32 Cloud TPU v3 cores for 1,000,000 steps with maximum sequence length 512 and batch size of 512.
In this setup, pre-training on MLM only takes around 3 days. Aditionally, the model has been further pre-trained on a second task (table entailment). See the original TAPAS [paper](https://www.aclweb.org/anthology/2020.acl-main.398/) and the [follow-up paper](https://www.aclweb.org/anthology/2020.findings-emnlp.27/) for more details.
The optimizer used is Adam with a learning rate of 5e-5, and a warmup
ratio of 0.01.
### BibTeX entry and citation info
```bibtex
@misc{herzig2020tapas,
title={TAPAS: Weakly Supervised Table Parsing via Pre-training},
author={Jonathan Herzig and Paweł Krzysztof Nowak and Thomas Müller and Francesco Piccinno and Julian Martin Eisenschlos},
year={2020},
eprint={2004.02349},
archivePrefix={arXiv},
primaryClass={cs.IR}
}
```
```bibtex
@misc{eisenschlos2020understanding,
title={Understanding tables with intermediate pre-training},
author={Julian Martin Eisenschlos and Syrine Krichene and Thomas Müller},
year={2020},
eprint={2010.00571},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
``` |
google/tapas-mini | google | 2021-11-29T10:11:56Z | 12 | 0 | transformers | [
"transformers",
"pytorch",
"tf",
"tapas",
"feature-extraction",
"TapasModel",
"en",
"arxiv:2004.02349",
"arxiv:2010.00571",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | feature-extraction | 2022-03-02T23:29:05Z | ---
language: en
tags:
- tapas
- TapasModel
license: apache-2.0
---
# TAPAS mini model
This model has 2 versions which can be used. The latest version, which is the default one, corresponds to the `tapas_inter_masklm_mini_reset` checkpoint of the [original Github repository](https://github.com/google-research/tapas).
This model was pre-trained on MLM and an additional step which the authors call intermediate pre-training. It uses relative position embeddings by default (i.e. resetting the position index at every cell of the table).
The other (non-default) version which can be used is the one with absolute position embeddings:
- `revision="no_reset"`, which corresponds to `tapas_inter_masklm_mini`
Disclaimer: The team releasing TAPAS did not write a model card for this model so this model card has been written by
the Hugging Face team and contributors.
## Model description
TAPAS is a BERT-like transformers model pretrained on a large corpus of English data from Wikipedia in a self-supervised fashion.
This means it was pretrained on the raw tables and associated texts only, with no humans labelling them in any way (which is why it
can use lots of publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a (flattened) table and associated context, the model randomly masks 15% of the words in
the input, then runs the entire (partially masked) sequence through the model. The model then has to predict the masked words.
This is different from traditional recurrent neural networks (RNNs) that usually see the words one after the other,
or from autoregressive models like GPT which internally mask the future tokens. It allows the model to learn a bidirectional
representation of a table and associated text.
- Intermediate pre-training: to encourage numerical reasoning on tables, the authors additionally pre-trained the model by creating
a balanced dataset of millions of syntactically created training examples. Here, the model must predict (classify) whether a sentence
is supported or refuted by the contents of a table. The training examples are created based on synthetic as well as counterfactual statements.
This way, the model learns an inner representation of the English language used in tables and associated texts, which can then be used
to extract features useful for downstream tasks such as answering questions about a table, or determining whether a sentence is entailed
or refuted by the contents of a table. Fine-tuning is done by adding one or more classification heads on top of the pre-trained model, and then
jointly train these randomly initialized classification heads with the base model on a downstream task.
## Intended uses & limitations
You can use the raw model for getting hidden representatons about table-question pairs, but it's mostly intended to be fine-tuned on a downstream task such as question answering or sequence classification. See the [model hub](https://huggingface.co/models?filter=tapas) to look for fine-tuned versions on a task that interests you.
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence [SEP] Flattened table [SEP]
```
### Pre-training
The model was pre-trained on 32 Cloud TPU v3 cores for 1,000,000 steps with maximum sequence length 512 and batch size of 512.
In this setup, pre-training on MLM only takes around 3 days. Aditionally, the model has been further pre-trained on a second task (table entailment). See the original TAPAS [paper](https://www.aclweb.org/anthology/2020.acl-main.398/) and the [follow-up paper](https://www.aclweb.org/anthology/2020.findings-emnlp.27/) for more details.
The optimizer used is Adam with a learning rate of 5e-5, and a warmup
ratio of 0.01.
### BibTeX entry and citation info
```bibtex
@misc{herzig2020tapas,
title={TAPAS: Weakly Supervised Table Parsing via Pre-training},
author={Jonathan Herzig and Paweł Krzysztof Nowak and Thomas Müller and Francesco Piccinno and Julian Martin Eisenschlos},
year={2020},
eprint={2004.02349},
archivePrefix={arXiv},
primaryClass={cs.IR}
}
```
```bibtex
@misc{eisenschlos2020understanding,
title={Understanding tables with intermediate pre-training},
author={Julian Martin Eisenschlos and Syrine Krichene and Thomas Müller},
year={2020},
eprint={2010.00571},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
``` |
google/tapas-tiny | google | 2021-11-29T10:01:08Z | 99 | 0 | transformers | [
"transformers",
"pytorch",
"tf",
"tapas",
"feature-extraction",
"TapasModel",
"en",
"arxiv:2004.02349",
"arxiv:2010.00571",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | feature-extraction | 2022-03-02T23:29:05Z | ---
language: en
tags:
- tapas
- TapasModel
license: apache-2.0
---
# TAPAS tiny model
This model has 2 versions which can be used. The latest version, which is the default one, corresponds to the `tapas_inter_masklm_tiny_reset` checkpoint of the [original Github repository](https://github.com/google-research/tapas).
This model was pre-trained on MLM and an additional step which the authors call intermediate pre-training. It uses relative position embeddings by default (i.e. resetting the position index at every cell of the table).
The other (non-default) version which can be used is the one with absolute position embeddings:
- `revision="no_reset"`, which corresponds to `tapas_inter_masklm_tiny`
Disclaimer: The team releasing TAPAS did not write a model card for this model so this model card has been written by
the Hugging Face team and contributors.
## Model description
TAPAS is a BERT-like transformers model pretrained on a large corpus of English data from Wikipedia in a self-supervised fashion.
This means it was pretrained on the raw tables and associated texts only, with no humans labelling them in any way (which is why it
can use lots of publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a (flattened) table and associated context, the model randomly masks 15% of the words in
the input, then runs the entire (partially masked) sequence through the model. The model then has to predict the masked words.
This is different from traditional recurrent neural networks (RNNs) that usually see the words one after the other,
or from autoregressive models like GPT which internally mask the future tokens. It allows the model to learn a bidirectional
representation of a table and associated text.
- Intermediate pre-training: to encourage numerical reasoning on tables, the authors additionally pre-trained the model by creating
a balanced dataset of millions of syntactically created training examples. Here, the model must predict (classify) whether a sentence
is supported or refuted by the contents of a table. The training examples are created based on synthetic as well as counterfactual statements.
This way, the model learns an inner representation of the English language used in tables and associated texts, which can then be used
to extract features useful for downstream tasks such as answering questions about a table, or determining whether a sentence is entailed
or refuted by the contents of a table. Fine-tuning is done by adding one or more classification heads on top of the pre-trained model, and then
jointly train these randomly initialized classification heads with the base model on a downstream task.
## Intended uses & limitations
You can use the raw model for getting hidden representatons about table-question pairs, but it's mostly intended to be fine-tuned on a downstream task such as question answering or sequence classification. See the [model hub](https://huggingface.co/models?filter=tapas) to look for fine-tuned versions on a task that interests you.
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence [SEP] Flattened table [SEP]
```
### Pre-training
The model was pre-trained on 32 Cloud TPU v3 cores for 1,000,000 steps with maximum sequence length 512 and batch size of 512.
In this setup, pre-training on MLM only takes around 3 days. Aditionally, the model has been further pre-trained on a second task (table entailment). See the original TAPAS [paper](https://www.aclweb.org/anthology/2020.acl-main.398/) and the [follow-up paper](https://www.aclweb.org/anthology/2020.findings-emnlp.27/) for more details.
The optimizer used is Adam with a learning rate of 5e-5, and a warmup
ratio of 0.01.
### BibTeX entry and citation info
```bibtex
@misc{herzig2020tapas,
title={TAPAS: Weakly Supervised Table Parsing via Pre-training},
author={Jonathan Herzig and Paweł Krzysztof Nowak and Thomas Müller and Francesco Piccinno and Julian Martin Eisenschlos},
year={2020},
eprint={2004.02349},
archivePrefix={arXiv},
primaryClass={cs.IR}
}
```
```bibtex
@misc{eisenschlos2020understanding,
title={Understanding tables with intermediate pre-training},
author={Julian Martin Eisenschlos and Syrine Krichene and Thomas Müller},
year={2020},
eprint={2010.00571},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
``` |
am4nsolanki/autonlp-text-hateful-memes-36789092 | am4nsolanki | 2021-11-28T22:35:30Z | 63 | 3 | transformers | [
"transformers",
"pytorch",
"distilbert",
"text-classification",
"autonlp",
"en",
"dataset:am4nsolanki/autonlp-data-text-hateful-memes",
"co2_eq_emissions",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text-classification | 2022-03-02T23:29:05Z | ---
tags: autonlp
language: en
widget:
- text: "I love AutoNLP 🤗"
datasets:
- am4nsolanki/autonlp-data-text-hateful-memes
co2_eq_emissions: 1.4280361775467445
---
# Model Trained Using AutoNLP
- Problem type: Binary Classification
- Model ID: 36789092
- CO2 Emissions (in grams): 1.4280361775467445
## Validation Metrics
- Loss: 0.5255328416824341
- Accuracy: 0.7666078777189889
- Precision: 0.6913123844731978
- Recall: 0.6192052980132451
- AUC: 0.7893359070795125
- F1: 0.6532751091703057
## Usage
You can use cURL to access this model:
```
$ curl -X POST -H "Authorization: Bearer YOUR_API_KEY" -H "Content-Type: application/json" -d '{"inputs": "I love AutoNLP"}' https://api-inference.huggingface.co/models/am4nsolanki/autonlp-text-hateful-memes-36789092
```
Or Python API:
```
from transformers import AutoModelForSequenceClassification, AutoTokenizer
model = AutoModelForSequenceClassification.from_pretrained("am4nsolanki/autonlp-text-hateful-memes-36789092", use_auth_token=True)
tokenizer = AutoTokenizer.from_pretrained("am4nsolanki/autonlp-text-hateful-memes-36789092", use_auth_token=True)
inputs = tokenizer("I love AutoNLP", return_tensors="pt")
outputs = model(**inputs)
``` |
Qinghui/autonlp-fake-covid-news-36769078 | Qinghui | 2021-11-28T19:41:07Z | 5 | 0 | transformers | [
"transformers",
"pytorch",
"roberta",
"text-classification",
"autonlp",
"unk",
"dataset:Qinghui/autonlp-data-fake-covid-news",
"co2_eq_emissions",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text-classification | 2022-03-02T23:29:04Z | ---
tags: autonlp
language: unk
widget:
- text: "I love AutoNLP 🤗"
datasets:
- Qinghui/autonlp-data-fake-covid-news
co2_eq_emissions: 23.42719853096565
---
# Model Trained Using AutoNLP
- Problem type: Binary Classification
- Model ID: 36769078
- CO2 Emissions (in grams): 23.42719853096565
## Validation Metrics
- Loss: 0.15959647297859192
- Accuracy: 0.9817757009345794
- Precision: 0.980411361410382
- Recall: 0.9813725490196078
- AUC: 0.9982379201680672
- F1: 0.9808917197452229
## Usage
You can use cURL to access this model:
```
$ curl -X POST -H "Authorization: Bearer YOUR_API_KEY" -H "Content-Type: application/json" -d '{"inputs": "I love AutoNLP"}' https://api-inference.huggingface.co/models/Qinghui/autonlp-fake-covid-news-36769078
```
Or Python API:
```
from transformers import AutoModelForSequenceClassification, AutoTokenizer
model = AutoModelForSequenceClassification.from_pretrained("Qinghui/autonlp-fake-covid-news-36769078", use_auth_token=True)
tokenizer = AutoTokenizer.from_pretrained("Qinghui/autonlp-fake-covid-news-36769078", use_auth_token=True)
inputs = tokenizer("I love AutoNLP", return_tensors="pt")
outputs = model(**inputs)
``` |
Matthijsvanhof/bert-base-dutch-cased-finetuned-mBERT | Matthijsvanhof | 2021-11-28T18:03:02Z | 5 | 0 | transformers | [
"transformers",
"pytorch",
"tensorboard",
"distilbert",
"token-classification",
"generated_from_trainer",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:04Z | ---
license: apache-2.0
tags:
- generated_from_trainer
metrics:
- precision
- recall
- f1
- accuracy
model-index:
- name: bert-base-dutch-cased-finetuned-mBERT
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# bert-base-dutch-cased-finetuned-mBERT
This model is a fine-tuned version of [distilbert-base-multilingual-cased](https://huggingface.co/distilbert-base-multilingual-cased) on an unknown dataset.
It achieves the following results on the evaluation set:
- Loss: 0.0898
- Precision: 0.7255
- Recall: 0.7255
- F1: 0.7255
- Accuracy: 0.9758
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 2
### Training results
| Training Loss | Epoch | Step | Validation Loss | Precision | Recall | F1 | Accuracy |
|:-------------:|:-----:|:----:|:---------------:|:---------:|:------:|:------:|:--------:|
| 0.1603 | 1.0 | 533 | 0.0928 | 0.6896 | 0.6962 | 0.6929 | 0.9742 |
| 0.0832 | 2.0 | 1066 | 0.0898 | 0.7255 | 0.7255 | 0.7255 | 0.9758 |
### Framework versions
- Transformers 4.12.5
- Pytorch 1.10.0+cu111
- Tokenizers 0.10.3
|
Alvenir/wav2vec2-base-da | Alvenir | 2021-11-28T11:35:11Z | 12 | 6 | transformers | [
"transformers",
"pytorch",
"wav2vec2",
"pretraining",
"speech",
"da",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04Z | ---
language: da
tags:
- speech
license: apache-2.0
---
# Wav2vec2-base for Danish
This wav2vec2-base model has been pretrained on ~1300 hours of danish speech data. The pretraining data consists of podcasts and audiobooks and is unfortunately not public available. However, we were allowed to distribute the pretrained model.
This model was pretrained on 16kHz sampled speech audio. When using the model, make sure to use speech audio sampled at 16kHz.
The pre-training was done using the fairseq library in January 2021.
It needs to be fine-tuned to perform speech recognition.
# Finetuning
In order to finetune the model to speech recognition, you can draw inspiration from this [notebook tutorial](https://colab.research.google.com/drive/1FjTsqbYKphl9kL-eILgUc-bl4zVThL8F) or [this blog post tutorial](https://huggingface.co/blog/fine-tune-xlsr-wav2vec2). |
aditi2222/t5-paraphrase | aditi2222 | 2021-11-28T07:35:16Z | 5 | 0 | transformers | [
"transformers",
"pytorch",
"t5",
"text2text-generation",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text2text-generation | 2022-03-02T23:29:05Z | T5 model
This is a sentence-transformers mode |
Matthijsvanhof/bert-base-dutch-cased-finetuned-NER8 | Matthijsvanhof | 2021-11-27T23:02:08Z | 5 | 0 | transformers | [
"transformers",
"pytorch",
"tensorboard",
"bert",
"token-classification",
"generated_from_trainer",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:04Z | ---
tags:
- generated_from_trainer
metrics:
- precision
- recall
- f1
- accuracy
model-index:
- name: bert-base-dutch-cased-finetuned-NER8
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# bert-base-dutch-cased-finetuned-NER8
This model is a fine-tuned version of [GroNLP/bert-base-dutch-cased](https://huggingface.co/GroNLP/bert-base-dutch-cased) on an unknown dataset.
It achieves the following results on the evaluation set:
- Loss: 0.1482
- Precision: 0.4716
- Recall: 0.4359
- F1: 0.4530
- Accuracy: 0.9569
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 2
### Training results
| Training Loss | Epoch | Step | Validation Loss | Precision | Recall | F1 | Accuracy |
|:-------------:|:-----:|:----:|:---------------:|:---------:|:------:|:------:|:--------:|
| No log | 1.0 | 68 | 0.1705 | 0.3582 | 0.3488 | 0.3535 | 0.9475 |
| No log | 2.0 | 136 | 0.1482 | 0.4716 | 0.4359 | 0.4530 | 0.9569 |
### Framework versions
- Transformers 4.12.5
- Pytorch 1.10.0+cu111
- Tokenizers 0.10.3
|
lgris/bp-voxforge1-xlsr | lgris | 2021-11-27T21:14:32Z | 3 | 0 | transformers | [
"transformers",
"pytorch",
"wav2vec2",
"automatic-speech-recognition",
"audio",
"speech",
"pt",
"portuguese-speech-corpus",
"PyTorch",
"dataset:common_voice",
"dataset:mls",
"dataset:cetuc",
"dataset:lapsbm",
"dataset:voxforge",
"dataset:tedx",
"dataset:sid",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | automatic-speech-recognition | 2022-03-02T23:29:05Z | ---
language: pt
datasets:
- common_voice
- mls
- cetuc
- lapsbm
- voxforge
- tedx
- sid
metrics:
- wer
tags:
- audio
- speech
- wav2vec2
- pt
- portuguese-speech-corpus
- automatic-speech-recognition
- speech
- PyTorch
license: apache-2.0
---
# voxforge1-xlsr: Wav2vec 2.0 with VoxForge Dataset
This is a the demonstration of a fine-tuned Wav2vec model for Brazilian Portuguese using the [VoxForge](http://www.voxforge.org/) dataset.
In this notebook the model is tested against other available Brazilian Portuguese datasets.
| Dataset | Train | Valid | Test |
|--------------------------------|-------:|------:|------:|
| CETUC | | -- | 5.4h |
| Common Voice | | -- | 9.5h |
| LaPS BM | | -- | 0.1h |
| MLS | | -- | 3.7h |
| Multilingual TEDx (Portuguese) | | -- | 1.8h |
| SID | | -- | 1.0h |
| VoxForge | 3.9h | -- | 0.1h |
| Total | 3.9h | -- | 21.6h |
#### Summary
| | CETUC | CV | LaPS | MLS | SID | TEDx | VF | AVG |
|----------------------|---------------|----------------|----------------|----------------|----------------|----------------|----------------|----------------|
| voxforge\_1 (demonstration below) | 0.468 | 0.608 | 0.503 | 0.505 | 0.717 | 0.731 | 0.561 | 0.584 |
| voxforge\_1 + 4-gram (demonstration below) | 0.322 | 0.471 | 0.356 | 0.378 | 0.586 | 0.637 | 0.428 | 0.454 |
## Demonstration
```python
MODEL_NAME = "lgris/voxforge1-xlsr"
```
### Imports and dependencies
```python
%%capture
!pip install torch==1.8.2+cu111 torchvision==0.9.2+cu111 torchaudio===0.8.2 -f https://download.pytorch.org/whl/lts/1.8/torch_lts.html
!pip install datasets
!pip install jiwer
!pip install transformers
!pip install soundfile
!pip install pyctcdecode
!pip install https://github.com/kpu/kenlm/archive/master.zip
```
```python
import jiwer
import torchaudio
from datasets import load_dataset, load_metric
from transformers import (
Wav2Vec2ForCTC,
Wav2Vec2Processor,
)
from pyctcdecode import build_ctcdecoder
import torch
import re
import sys
```
### Helpers
```python
chars_to_ignore_regex = '[\,\?\.\!\;\:\"]' # noqa: W605
def map_to_array(batch):
speech, _ = torchaudio.load(batch["path"])
batch["speech"] = speech.squeeze(0).numpy()
batch["sampling_rate"] = 16_000
batch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower().replace("’", "'")
batch["target"] = batch["sentence"]
return batch
```
```python
def calc_metrics(truths, hypos):
wers = []
mers = []
wils = []
for t, h in zip(truths, hypos):
try:
wers.append(jiwer.wer(t, h))
mers.append(jiwer.mer(t, h))
wils.append(jiwer.wil(t, h))
except: # Empty string?
pass
wer = sum(wers)/len(wers)
mer = sum(mers)/len(mers)
wil = sum(wils)/len(wils)
return wer, mer, wil
```
```python
def load_data(dataset):
data_files = {'test': f'{dataset}/test.csv'}
dataset = load_dataset('csv', data_files=data_files)["test"]
return dataset.map(map_to_array)
```
### Model
```python
class STT:
def __init__(self,
model_name,
device='cuda' if torch.cuda.is_available() else 'cpu',
lm=None):
self.model_name = model_name
self.model = Wav2Vec2ForCTC.from_pretrained(model_name).to(device)
self.processor = Wav2Vec2Processor.from_pretrained(model_name)
self.vocab_dict = self.processor.tokenizer.get_vocab()
self.sorted_dict = {
k.lower(): v for k, v in sorted(self.vocab_dict.items(),
key=lambda item: item[1])
}
self.device = device
self.lm = lm
if self.lm:
self.lm_decoder = build_ctcdecoder(
list(self.sorted_dict.keys()),
self.lm
)
def batch_predict(self, batch):
features = self.processor(batch["speech"],
sampling_rate=batch["sampling_rate"][0],
padding=True,
return_tensors="pt")
input_values = features.input_values.to(self.device)
attention_mask = features.attention_mask.to(self.device)
with torch.no_grad():
logits = self.model(input_values, attention_mask=attention_mask).logits
if self.lm:
logits = logits.cpu().numpy()
batch["predicted"] = []
for sample_logits in logits:
batch["predicted"].append(self.lm_decoder.decode(sample_logits))
else:
pred_ids = torch.argmax(logits, dim=-1)
batch["predicted"] = self.processor.batch_decode(pred_ids)
return batch
```
### Download datasets
```python
%%capture
!gdown --id 1HFECzIizf-bmkQRLiQD0QVqcGtOG5upI
!mkdir bp_dataset
!unzip bp_dataset -d bp_dataset/
```
### Tests
```python
stt = STT(MODEL_NAME)
```
#### CETUC
```python
ds = load_data('cetuc_dataset')
result = ds.map(stt.batch_predict, batched=True, batch_size=8)
wer, mer, wil = calc_metrics(result["sentence"], result["predicted"])
print("CETUC WER:", wer)
```
CETUC WER: 0.4684840205331983
#### Common Voice
```python
ds = load_data('commonvoice_dataset')
result = ds.map(stt.batch_predict, batched=True, batch_size=8)
wer, mer, wil = calc_metrics(result["sentence"], result["predicted"])
print("CV WER:", wer)
```
CV WER: 0.6080167359840954
#### LaPS
```python
ds = load_data('lapsbm_dataset')
result = ds.map(stt.batch_predict, batched=True, batch_size=8)
wer, mer, wil = calc_metrics(result["sentence"], result["predicted"])
print("Laps WER:", wer)
```
Laps WER: 0.5037468434343434
#### MLS
```python
ds = load_data('mls_dataset')
result = ds.map(stt.batch_predict, batched=True, batch_size=8)
wer, mer, wil = calc_metrics(result["sentence"], result["predicted"])
print("MLS WER:", wer)
```
MLS WER: 0.505595213971485
#### SID
```python
ds = load_data('sid_dataset')
result = ds.map(stt.batch_predict, batched=True, batch_size=8)
wer, mer, wil = calc_metrics(result["sentence"], result["predicted"])
print("Sid WER:", wer)
```
Sid WER: 0.7177723323755854
#### TEDx
```python
ds = load_data('tedx_dataset')
result = ds.map(stt.batch_predict, batched=True, batch_size=8)
wer, mer, wil = calc_metrics(result["sentence"], result["predicted"])
print("TEDx WER:", wer)
```
TEDx WER: 0.7309431974873112
#### VoxForge
```python
ds = load_data('voxforge_dataset')
result = ds.map(stt.batch_predict, batched=True, batch_size=8)
wer, mer, wil = calc_metrics(result["sentence"], result["predicted"])
print("VoxForge WER:", wer)
```
VoxForge WER: 0.5613906926406929
### Tests with LM
```python
# !find -type f -name "*.wav" -delete
!rm -rf ~/.cache
!gdown --id 1GJIKseP5ZkTbllQVgOL98R4yYAcIySFP # trained with wikipedia
stt = STT(MODEL_NAME, lm='pt-BR-wiki.word.4-gram.arpa')
# !gdown --id 1dLFldy7eguPtyJj5OAlI4Emnx0BpFywg # trained with bp
# stt = STT(MODEL_NAME, lm='pt-BR.word.4-gram.arpa')
```
#### CETUC
```python
ds = load_data('cetuc_dataset')
result = ds.map(stt.batch_predict, batched=True, batch_size=8)
wer, mer, wil = calc_metrics(result["sentence"], result["predicted"])
print("CETUC WER:", wer)
```
CETUC WER: 0.32184971297675896
#### Common Voice
```python
ds = load_data('commonvoice_dataset')
result = ds.map(stt.batch_predict, batched=True, batch_size=8)
wer, mer, wil = calc_metrics(result["sentence"], result["predicted"])
print("CV WER:", wer)
```
CV WER: 0.4707820098981609
#### LaPS
```python
ds = load_data('lapsbm_dataset')
result = ds.map(stt.batch_predict, batched=True, batch_size=8)
wer, mer, wil = calc_metrics(result["sentence"], result["predicted"])
print("Laps WER:", wer)
```
Laps WER: 0.356227904040404
#### MLS
```python
ds = load_data('mls_dataset')
result = ds.map(stt.batch_predict, batched=True, batch_size=8)
wer, mer, wil = calc_metrics(result["sentence"], result["predicted"])
print("MLS WER:", wer)
```
MLS WER: 0.3786376653384398
#### SID
```python
ds = load_data('sid_dataset')
result = ds.map(stt.batch_predict, batched=True, batch_size=8)
wer, mer, wil = calc_metrics(result["sentence"], result["predicted"])
print("Sid WER:", wer)
```
Sid WER: 0.5864959640811857
#### TEDx
```python
ds = load_data('tedx_dataset')
result = ds.map(stt.batch_predict, batched=True, batch_size=8)
wer, mer, wil = calc_metrics(result["sentence"], result["predicted"])
print("TEDx WER:", wer)
```
TEDx WER: 0.6368727228726417
#### VoxForge
```python
ds = load_data('voxforge_dataset')
result = ds.map(stt.batch_predict, batched=True, batch_size=8)
wer, mer, wil = calc_metrics(result["sentence"], result["predicted"])
print("VoxForge WER:", wer)
```
VoxForge WER: 0.4279924242424241
|
baby-oogway/wav2vec2-timit_asr-oogway | baby-oogway | 2021-11-27T20:14:26Z | 4 | 0 | transformers | [
"transformers",
"pytorch",
"tensorboard",
"wav2vec2",
"automatic-speech-recognition",
"generated_from_trainer",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | automatic-speech-recognition | 2022-03-02T23:29:05Z | ---
license: apache-2.0
tags:
- generated_from_trainer
model-index:
- name: wav2vec2-timit_asr-oogway
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# wav2vec2-timit_asr-oogway
This model is a fine-tuned version of [OthmaneJ/distil-wav2vec2](https://huggingface.co/OthmaneJ/distil-wav2vec2) on the None dataset.
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 0.0001
- train_batch_size: 32
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 1000
- num_epochs: 30
- mixed_precision_training: Native AMP
### Framework versions
- Transformers 4.11.3
- Pytorch 1.10.0+cu111
- Datasets 1.13.3
- Tokenizers 0.10.3
|
world-wide/sent-sci-irrelevance | world-wide | 2021-11-27T14:16:04Z | 5 | 1 | transformers | [
"transformers",
"pytorch",
"bert",
"text-classification",
"autonlp",
"en",
"dataset:bozelosp/autonlp-data-sci-relevance",
"co2_eq_emissions",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text-classification | 2022-03-02T23:29:05Z | ---
tags: autonlp
language: en
widget:
- text: "I love AutoNLP 🤗"
datasets:
- bozelosp/autonlp-data-sci-relevance
co2_eq_emissions: 3.667033499762825
---
# Model Trained Using AutoNLP
- Problem type: Binary Classification
- Model ID: 33199029
- CO2 Emissions (in grams): 3.667033499762825
## Validation Metrics
- Loss: 0.32653310894966125
- Accuracy: 0.9133333333333333
- Precision: 0.9005847953216374
- Recall: 0.9447852760736196
- AUC: 0.9532488468944517
- F1: 0.9221556886227544
## Usage
You can use cURL to access this model:
```
$ curl -X POST -H "Authorization: Bearer YOUR_API_KEY" -H "Content-Type: application/json" -d '{"inputs": "I love AutoNLP"}' https://api-inference.huggingface.co/models/bozelosp/autonlp-sci-relevance-33199029
```
Or Python API:
```
from transformers import AutoModelForSequenceClassification, AutoTokenizer
model = AutoModelForSequenceClassification.from_pretrained("bozelosp/autonlp-sci-relevance-33199029", use_auth_token=True)
tokenizer = AutoTokenizer.from_pretrained("bozelosp/autonlp-sci-relevance-33199029", use_auth_token=True)
inputs = tokenizer("I love AutoNLP", return_tensors="pt")
outputs = model(**inputs)
``` |
tiagohatta/opus-mt-de-en-finetuned-de-to-en-first | tiagohatta | 2021-11-27T13:04:18Z | 3 | 0 | transformers | [
"transformers",
"pytorch",
"tensorboard",
"marian",
"text2text-generation",
"generated_from_trainer",
"dataset:wmt16",
"license:apache-2.0",
"model-index",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text2text-generation | 2022-03-02T23:29:05Z | ---
license: apache-2.0
tags:
- generated_from_trainer
datasets:
- wmt16
metrics:
- bleu
model-index:
- name: opus-mt-de-en-finetuned-de-to-en-first
results:
- task:
name: Sequence-to-sequence Language Modeling
type: text2text-generation
dataset:
name: wmt16
type: wmt16
args: de-en
metrics:
- name: Bleu
type: bleu
value: 39.8122
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# opus-mt-de-en-finetuned-de-to-en-first
This model is a fine-tuned version of [Helsinki-NLP/opus-mt-de-en](https://huggingface.co/Helsinki-NLP/opus-mt-de-en) on the wmt16 dataset.
It achieves the following results on the evaluation set:
- Loss: 1.1465
- Bleu: 39.8122
- Gen Len: 25.579
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 1
- mixed_precision_training: Native AMP
### Training results
| Training Loss | Epoch | Step | Validation Loss | Bleu | Gen Len |
|:-------------:|:-----:|:----:|:---------------:|:-------:|:-------:|
| No log | 1.0 | 63 | 1.1465 | 39.8122 | 25.579 |
### Framework versions
- Transformers 4.12.5
- Pytorch 1.10.0+cu111
- Datasets 1.16.1
- Tokenizers 0.10.3
|
dandelin/vilt-b32-mlm-itm | dandelin | 2021-11-27T10:13:10Z | 686 | 2 | transformers | [
"transformers",
"pytorch",
"vilt",
"arxiv:2102.03334",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:05Z | ---
license: apache-2.0
tags:
---
# Vision-and-Language Transformer (ViLT), pre-trained only
Vision-and-Language Transformer (ViLT) model pre-trained on GCC+SBU+COCO+VG (200k steps). It was introduced in the paper [ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision](https://arxiv.org/abs/2102.03334) by Kim et al. and first released in [this repository](https://github.com/dandelin/ViLT).
Disclaimer: The team releasing ViLT did not write a model card for this model so this model card has been written by the Hugging Face team.
## Model description
(to do)
## Intended uses & limitations
You can use the raw model for visual question answering.
### How to use
(to do)
## Training data
(to do)
## Training procedure
### Preprocessing
(to do)
### Pretraining
(to do)
## Evaluation results
(to do)
### BibTeX entry and citation info
```bibtex
@misc{kim2021vilt,
title={ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision},
author={Wonjae Kim and Bokyung Son and Ildoo Kim},
year={2021},
eprint={2102.03334},
archivePrefix={arXiv},
primaryClass={stat.ML}
}
``` |
cambridgeltl/trans-encoder-bi-simcse-bert-base | cambridgeltl | 2021-11-26T18:26:34Z | 7 | 0 | transformers | [
"transformers",
"pytorch",
"bert",
"feature-extraction",
"arxiv:2109.13059",
"text-embeddings-inference",
"endpoints_compatible",
"region:us"
] | feature-extraction | 2022-03-02T23:29:05Z | ---
language: en
tags:
- sentence-embeddings
- sentence-similarity
- dual-encoder
### cambridgeltl/trans-encoder-bi-simcse-bert-base
An unsupervised sentence encoder (bi-encoder) proposed by [Liu et al. (2021)](https://arxiv.org/pdf/2109.13059.pdf). The model is trained with unlabelled sentence pairs sampled from STS2012-2016, STS-b, and SICK-R, using [princeton-nlp/unsup-simcse-bert-base-uncased](https://huggingface.co/princeton-nlp/unsup-simcse-bert-base-uncased) as the base model. Please use `[CLS]` (before pooler) as the representation of the input.
### Citation
```bibtex
@article{liu2021trans,
title={Trans-Encoder: Unsupervised sentence-pair modelling through self-and mutual-distillations},
author={Liu, Fangyu and Jiao, Yunlong and Massiah, Jordan and Yilmaz, Emine and Havrylov, Serhii},
journal={arXiv preprint arXiv:2109.13059},
year={2021}
}
```
|
arnav7633/DialoGPT-medium-tony_stark | arnav7633 | 2021-11-26T17:06:17Z | 4 | 0 | transformers | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"conversational",
"en",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text-generation | 2022-03-02T23:29:05Z | ---
language:
- en
tags:
- conversational
license: MIT
---
**A casual chatbot**
This is a dialogpt medium fine tuned to talk like Tony Stark, Currently its only trained upon the script of Iron man 3 |
huggingtweets/insharamin-prathkum-saviomartin7 | huggingtweets | 2021-11-26T10:18:51Z | 6 | 0 | transformers | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"huggingtweets",
"en",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text-generation | 2022-03-02T23:29:05Z | ---
language: en
thumbnail: http://www.huggingtweets.com/insharamin-prathkum-saviomartin7/1637920907734/predictions.png
tags:
- huggingtweets
widget:
- text: "My dream is"
---
<div class="inline-flex flex-col" style="line-height: 1.5;">
<div class="flex">
<div
style="display:inherit; margin-left: 4px; margin-right: 4px; width: 92px; height:92px; border-radius: 50%; background-size: cover; background-image: url('https://pbs.twimg.com/profile_images/1418652395119153153/dvMUbHmM_400x400.jpg')">
</div>
<div
style="display:inherit; margin-left: 4px; margin-right: 4px; width: 92px; height:92px; border-radius: 50%; background-size: cover; background-image: url('https://pbs.twimg.com/profile_images/1449364913890074627/SNmSlTYD_400x400.jpg')">
</div>
<div
style="display:inherit; margin-left: 4px; margin-right: 4px; width: 92px; height:92px; border-radius: 50%; background-size: cover; background-image: url('https://pbs.twimg.com/profile_images/1450840619132260357/r9rdJtIp_400x400.jpg')">
</div>
</div>
<div style="text-align: center; margin-top: 3px; font-size: 16px; font-weight: 800">🤖 AI CYBORG 🤖</div>
<div style="text-align: center; font-size: 16px; font-weight: 800">Pratham & Insha & Savio Martin ⚡️</div>
<div style="text-align: center; font-size: 14px;">@insharamin-prathkum-saviomartin7</div>
</div>
I was made with [huggingtweets](https://github.com/borisdayma/huggingtweets).
Create your own bot based on your favorite user with [the demo](https://colab.research.google.com/github/borisdayma/huggingtweets/blob/master/huggingtweets-demo.ipynb)!
## How does it work?
The model uses the following pipeline.
To understand how the model was developed, check the [W&B report](https://wandb.ai/wandb/huggingtweets/reports/HuggingTweets-Train-a-Model-to-Generate-Tweets--VmlldzoxMTY5MjI).
## Training data
The model was trained on tweets from Pratham & Insha & Savio Martin ⚡️.
| Data | Pratham | Insha | Savio Martin ⚡️ |
| --- | --- | --- | --- |
| Tweets downloaded | 3246 | 3249 | 3249 |
| Retweets | 461 | 24 | 118 |
| Short tweets | 317 | 457 | 201 |
| Tweets kept | 2468 | 2768 | 2930 |
[Explore the data](https://wandb.ai/wandb/huggingtweets/runs/o7jfvmhp/artifacts), which is tracked with [W&B artifacts](https://docs.wandb.com/artifacts) at every step of the pipeline.
## Training procedure
The model is based on a pre-trained [GPT-2](https://huggingface.co/gpt2) which is fine-tuned on @insharamin-prathkum-saviomartin7's tweets.
Hyperparameters and metrics are recorded in the [W&B training run](https://wandb.ai/wandb/huggingtweets/runs/p2md0wva) for full transparency and reproducibility.
At the end of training, [the final model](https://wandb.ai/wandb/huggingtweets/runs/p2md0wva/artifacts) is logged and versioned.
## How to use
You can use this model directly with a pipeline for text generation:
```python
from transformers import pipeline
generator = pipeline('text-generation',
model='huggingtweets/insharamin-prathkum-saviomartin7')
generator("My dream is", num_return_sequences=5)
```
## Limitations and bias
The model suffers from [the same limitations and bias as GPT-2](https://huggingface.co/gpt2#limitations-and-bias).
In addition, the data present in the user's tweets further affects the text generated by the model.
## About
*Built by Boris Dayma*
[](https://twitter.com/intent/follow?screen_name=borisdayma)
For more details, visit the project repository.
[](https://github.com/borisdayma/huggingtweets)
|
Osiris/emotion_classifier | Osiris | 2021-11-26T07:57:27Z | 9 | 1 | transformers | [
"transformers",
"pytorch",
"roberta",
"text-classification",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text-classification | 2022-03-02T23:29:04Z | ### Introduction:
This model belongs to text-classification. You can determine the emotion behind a sentence.
### Label Explaination:
LABEL_0: Positive (have positive emotion)
LABEL_1: Negative (have negative emotion)
### Usage:
```python
>>> from transformers import pipeline
>>> ec = pipeline('text-classification', model='Osiris/emotion_classifier')
>>> ec("Hello, I'm a good model.")
```
### Accuracy:
We reach 83.82% for validation dataset, and 84.42% for test dataset. |
mrm8488/legalectra-base-spanish | mrm8488 | 2021-11-25T20:42:48Z | 17 | 3 | transformers | [
"transformers",
"pytorch",
"electra",
"pretraining",
"Spanish",
"Electra",
"Legal",
"es",
"dataset:Spanish-legal-corpora",
"arxiv:1406.2661",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:05Z | ---
language: es
tags:
- Spanish
- Electra
- Legal
datasets:
- Spanish-legal-corpora
---
## LEGALECTRA ⚖️
**LEGALECTRA** (base) is an Electra like model (discriminator in this case) trained on [A collection of corpora of Spanish legal domain](https://zenodo.org/record/5495529#.YZItp3vMLJw).
As mentioned in the original [paper](https://openreview.net/pdf?id=r1xMH1BtvB):
**ELECTRA** is a new method for self-supervised language representation learning. It can be used to pre-train transformer networks using relatively little compute. ELECTRA models are trained to distinguish "real" input tokens vs "fake" input tokens generated by another neural network, similar to the discriminator of a [GAN](https://arxiv.org/pdf/1406.2661.pdf). At small scale, ELECTRA achieves strong results even when trained on a single GPU. At large scale, ELECTRA achieves state-of-the-art results on the [SQuAD 2.0](https://rajpurkar.github.io/SQuAD-explorer/) dataset.
For a detailed description and experimental results, please refer the paper [ELECTRA: Pre-training Text Encoders as Discriminators Rather Than Generators](https://openreview.net/pdf?id=r1xMH1BtvB).
## Training details
TBA
## Model details ⚙
|Name| # Value|
|-----|--------|
|Layers| 12 |
|Hidden | 768 |
|Params| 110M |
## Evaluation metrics (for discriminator) 🧾
|Metric | # Score |
|-------|---------|
|Accuracy| 0.941|
|AUC | 0.794|
|Precision| |
## Benchmarks 🔨
WIP 🚧
## How to use the discriminator in `transformers`
TBA
## Acknowledgments
TBA
> Created by [Manuel Romero/@mrm8488](https://twitter.com/mrm8488)
> Made with <span style="color: #e25555;">♥</span> in Spain |
balamurugan1603/bert-finetuned-ner | balamurugan1603 | 2021-11-25T17:00:00Z | 19 | 1 | transformers | [
"transformers",
"pytorch",
"tf",
"bert",
"token-classification",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:05Z | # Named Entity Recognition using Transformers
This is a Fine-tuned version of BERT using HuggingFace transformers to perform Named Entity Recognition on Text data. BERT is a state-of-the-art model with attention mechanism as underlying architecture trained with masked-language-modeling and next-sentence-prediction objectives, used for various tasks including Question answering systems, Text Summarization, etc... which can also perform token classification tasks such as NER with great performance.
# Dataset
**CoNLL-2003** :
The shared task of CoNLL-2003 concerns language-independent named entity recognition. We will concentrate on four types of named entities: persons, locations, organizations, and names of miscellaneous entities that do not belong to the previous three groups.<br><br>
**Link** : https://huggingface.co/datasets/conll2003
# Using this fine-tuned version
From python, download the whole pipeline and use it instantly using the following code :
```
from transformers import pipeline
# Loading the pipeline from hub
# Pipeline handles the preprocessing and post processing steps
model_checkpoint = "balamurugan1603/bert-finetuned-ner"
namedEntityRecogniser = pipeline(
"token-classification", model=model_checkpoint, aggregation_strategy="simple"
)
```
Reference for using this pipeline to find NER tags can be found in this <a href="https://github.com/balamurugan1603/Named-Entity-Recognition-using-Tranformers/blob/main/named-entity-recognition-using-transfer-learning.ipynb">notebook</a>.
|
abdouaziiz/bert-base-wolof | abdouaziiz | 2021-11-25T16:35:19Z | 16 | 1 | transformers | [
"transformers",
"pytorch",
"bert",
"fill-mask",
"language-model",
"wo",
"wolof",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | fill-mask | 2022-03-02T23:29:05Z | ---
language: wo
tags:
- bert
- language-model
- wo
- wolof
---
# Soraberta: Unsupervised Language Model Pre-training for Wolof
**bert-base-wolof** is pretrained bert-base model on wolof language .
## Soraberta models
| Model name | Number of layers | Attention Heads | Embedding Dimension | Total Parameters |
| :------: | :---: | :---: | :---: | :---: |
| `bert-base` | 6 | 12 | 514 | 56931622 M |
## Using Soraberta with Hugging Face's Transformers
```python
>>> from transformers import pipeline
>>> unmasker = pipeline('fill-mask', model='abdouaziiz/bert-base-wolof')
>>> unmasker("kuy yoot du [MASK].")
[{'sequence': '[CLS] kuy yoot du seqet. [SEP]',
'score': 0.09505125880241394,
'token': 13578},
{'sequence': '[CLS] kuy yoot du daw. [SEP]',
'score': 0.08882280439138412,
'token': 679},
{'sequence': '[CLS] kuy yoot du yoot. [SEP]',
'score': 0.057790059596300125,
'token': 5117},
{'sequence': '[CLS] kuy yoot du seqat. [SEP]',
'score': 0.05671025067567825,
'token': 4992},
{'sequence': '[CLS] kuy yoot du yaqu. [SEP]',
'score': 0.0469999685883522,
'token': 1735}]
```
## Training data
The data sources are [Bible OT](http://biblewolof.com/) , [WOLOF-ONLINE](http://www.wolof-online.com/)
[ALFFA_PUBLIC](https://github.com/getalp/ALFFA_PUBLIC/tree/master/ASR/WOLOF)
## Contact
Please contact [email protected] for any question, feedback or request. |
huggingtweets/profdemirtas | huggingtweets | 2021-11-25T12:37:19Z | 4 | 0 | transformers | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"huggingtweets",
"en",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text-generation | 2022-03-02T23:29:05Z | ---
language: en
thumbnail: http://www.huggingtweets.com/profdemirtas/1637843815628/predictions.png
tags:
- huggingtweets
widget:
- text: "My dream is"
---
<div class="inline-flex flex-col" style="line-height: 1.5;">
<div class="flex">
<div
style="display:inherit; margin-left: 4px; margin-right: 4px; width: 92px; height:92px; border-radius: 50%; background-size: cover; background-image: url('https://pbs.twimg.com/profile_images/1374615485573165057/-AzXW69D_400x400.jpg')">
</div>
<div
style="display:none; margin-left: 4px; margin-right: 4px; width: 92px; height:92px; border-radius: 50%; background-size: cover; background-image: url('')">
</div>
<div
style="display:none; margin-left: 4px; margin-right: 4px; width: 92px; height:92px; border-radius: 50%; background-size: cover; background-image: url('')">
</div>
</div>
<div style="text-align: center; margin-top: 3px; font-size: 16px; font-weight: 800">🤖 AI BOT 🤖</div>
<div style="text-align: center; font-size: 16px; font-weight: 800">Özgür Demirtaş</div>
<div style="text-align: center; font-size: 14px;">@profdemirtas</div>
</div>
I was made with [huggingtweets](https://github.com/borisdayma/huggingtweets).
Create your own bot based on your favorite user with [the demo](https://colab.research.google.com/github/borisdayma/huggingtweets/blob/master/huggingtweets-demo.ipynb)!
## How does it work?
The model uses the following pipeline.
To understand how the model was developed, check the [W&B report](https://wandb.ai/wandb/huggingtweets/reports/HuggingTweets-Train-a-Model-to-Generate-Tweets--VmlldzoxMTY5MjI).
## Training data
The model was trained on tweets from Özgür Demirtaş.
| Data | Özgür Demirtaş |
| --- | --- |
| Tweets downloaded | 3205 |
| Retweets | 930 |
| Short tweets | 526 |
| Tweets kept | 1749 |
[Explore the data](https://wandb.ai/wandb/huggingtweets/runs/1ijpxe11/artifacts), which is tracked with [W&B artifacts](https://docs.wandb.com/artifacts) at every step of the pipeline.
## Training procedure
The model is based on a pre-trained [GPT-2](https://huggingface.co/gpt2) which is fine-tuned on @profdemirtas's tweets.
Hyperparameters and metrics are recorded in the [W&B training run](https://wandb.ai/wandb/huggingtweets/runs/1pvxmqhr) for full transparency and reproducibility.
At the end of training, [the final model](https://wandb.ai/wandb/huggingtweets/runs/1pvxmqhr/artifacts) is logged and versioned.
## How to use
You can use this model directly with a pipeline for text generation:
```python
from transformers import pipeline
generator = pipeline('text-generation',
model='huggingtweets/profdemirtas')
generator("My dream is", num_return_sequences=5)
```
## Limitations and bias
The model suffers from [the same limitations and bias as GPT-2](https://huggingface.co/gpt2#limitations-and-bias).
In addition, the data present in the user's tweets further affects the text generated by the model.
## About
*Built by Boris Dayma*
[](https://twitter.com/intent/follow?screen_name=borisdayma)
For more details, visit the project repository.
[](https://github.com/borisdayma/huggingtweets)
|
mbeukman/xlm-roberta-base-finetuned-yoruba-finetuned-ner-yoruba | mbeukman | 2021-11-25T09:05:18Z | 14 | 1 | transformers | [
"transformers",
"pytorch",
"xlm-roberta",
"token-classification",
"NER",
"yo",
"dataset:masakhaner",
"arxiv:2103.11811",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:05Z | ---
language:
- yo
tags:
- NER
datasets:
- masakhaner
metrics:
- f1
- precision
- recall
widget:
- text: "Kò sí ẹ̀rí tí ó fi ẹsẹ̀ rinlẹ̀ ."
---
# xlm-roberta-base-finetuned-yoruba-finetuned-ner-yoruba
This is a token classification (specifically NER) model that fine-tuned [xlm-roberta-base-finetuned-yoruba](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-yoruba) on the [MasakhaNER](https://arxiv.org/abs/2103.11811) dataset, specifically the Yoruba part.
More information, and other similar models can be found in the [main Github repository](https://github.com/Michael-Beukman/NERTransfer).
## About
This model is transformer based and was fine-tuned on the MasakhaNER dataset. It is a named entity recognition dataset, containing mostly news articles in 10 different African languages.
The model was fine-tuned for 50 epochs, with a maximum sequence length of 200, 32 batch size, 5e-5 learning rate. This process was repeated 5 times (with different random seeds), and this uploaded model performed the best out of those 5 seeds (aggregate F1 on test set).
This model was fine-tuned by me, Michael Beukman while doing a project at the University of the Witwatersrand, Johannesburg. This is version 1, as of 20 November 2021.
This model is licensed under the [Apache License, Version 2.0](https://www.apache.org/licenses/LICENSE-2.0).
### Contact & More information
For more information about the models, including training scripts, detailed results and further resources, you can visit the the [main Github repository](https://github.com/Michael-Beukman/NERTransfer). You can contact me by filing an issue on this repository.
### Training Resources
In the interest of openness, and reporting resources used, we list here how long the training process took, as well as what the minimum resources would be to reproduce this. Fine-tuning each model on the NER dataset took between 10 and 30 minutes, and was performed on a NVIDIA RTX3090 GPU. To use a batch size of 32, at least 14GB of GPU memory was required, although it was just possible to fit these models in around 6.5GB's of VRAM when using a batch size of 1.
## Data
The train, evaluation and test datasets were taken directly from the MasakhaNER [Github](https://github.com/masakhane-io/masakhane-ner) repository, with minimal to no preprocessing, as the original dataset is already of high quality.
The motivation for the use of this data is that it is the "first large, publicly available, high quality dataset for named entity recognition (NER) in ten African languages" ([source](https://arxiv.org/pdf/2103.11811.pdf)). The high-quality data, as well as the groundwork laid by the paper introducing it are some more reasons why this dataset was used. For evaluation, the dedicated test split was used, which is from the same distribution as the training data, so this model may not generalise to other distributions, and further testing would need to be done to investigate this. The exact distribution of the data is covered in detail [here](https://arxiv.org/abs/2103.11811).
## Intended Use
This model are intended to be used for NLP research into e.g. interpretability or transfer learning. Using this model in production is not supported, as generalisability and downright performance is limited. In particular, this is not designed to be used in any important downstream task that could affect people, as harm could be caused by the limitations of the model, described next.
## Limitations
This model was only trained on one (relatively small) dataset, covering one task (NER) in one domain (news articles) and in a set span of time. The results may not generalise, and the model may perform badly, or in an unfair / biased way if used on other tasks. Although the purpose of this project was to investigate transfer learning, the performance on languages that the model was not trained for does suffer.
Because this model used xlm-roberta-base as its starting point (potentially with domain adaptive fine-tuning on specific languages), this model's limitations can also apply here. These can include being biased towards the hegemonic viewpoint of most of its training data, being ungrounded and having subpar results on other languages (possibly due to unbalanced training data).
As [Adelani et al. (2021)](https://arxiv.org/abs/2103.11811) showed, the models in general struggled with entities that were either longer than 3 words and entities that were not contained in the training data. This could bias the models towards not finding, e.g. names of people that have many words, possibly leading to a misrepresentation in the results. Similarly, names that are uncommon, and may not have been found in the training data (due to e.g. different languages) would also be predicted less often.
Additionally, this model has not been verified in practice, and other, more subtle problems may become prevalent if used without any verification that it does what it is supposed to.
### Privacy & Ethical Considerations
The data comes from only publicly available news sources, the only available data should cover public figures and those that agreed to be reported on. See the original MasakhaNER paper for more details.
No explicit ethical considerations or adjustments were made during fine-tuning of this model.
## Metrics
The language adaptive models achieve (mostly) superior performance over starting with xlm-roberta-base. Our main metric was the aggregate F1 score for all NER categories.
These metrics are on the test set for MasakhaNER, so the data distribution is similar to the training set, so these results do not directly indicate how well these models generalise.
We do find large variation in transfer results when starting from different seeds (5 different seeds were tested), indicating that the fine-tuning process for transfer might be unstable.
The metrics used were chosen to be consistent with previous work, and to facilitate research. Other metrics may be more appropriate for other purposes.
## Caveats and Recommendations
In general, this model performed worse on the 'date' category compared to others, so if dates are a critical factor, then that might need to be taken into account and addressed, by for example collecting and annotating more data.
## Model Structure
Here are some performance details on this specific model, compared to others we trained.
All of these metrics were calculated on the test set, and the seed was chosen that gave the best overall F1 score. The first three result columns are averaged over all categories, and the latter 4 provide performance broken down by category.
This model can predict the following label for a token ([source](https://huggingface.co/Davlan/xlm-roberta-large-masakhaner)):
Abbreviation|Description
-|-
O|Outside of a named entity
B-DATE |Beginning of a DATE entity right after another DATE entity
I-DATE |DATE entity
B-PER |Beginning of a person’s name right after another person’s name
I-PER |Person’s name
B-ORG |Beginning of an organisation right after another organisation
I-ORG |Organisation
B-LOC |Beginning of a location right after another location
I-LOC |Location
| Model Name | Staring point | Evaluation / Fine-tune Language | F1 | Precision | Recall | F1 (DATE) | F1 (LOC) | F1 (ORG) | F1 (PER) |
| -------------------------------------------------- | -------------------- | -------------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- |
| [xlm-roberta-base-finetuned-yoruba-finetuned-ner-yoruba](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-yoruba-finetuned-ner-yoruba) (This model) | [yor](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-yoruba) | yor | 83.68 | 79.92 | 87.82 | 78.00 | 86.00 | 74.00 | 92.00 |
| [xlm-roberta-base-finetuned-swahili-finetuned-ner-yoruba](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-yoruba) | [swa](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-swahili) | yor | 80.29 | 78.34 | 82.35 | 77.00 | 82.00 | 73.00 | 86.00 |
| [xlm-roberta-base-finetuned-ner-yoruba](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-ner-yoruba) | [base](https://huggingface.co/xlm-roberta-base) | yor | 78.22 | 77.21 | 79.26 | 77.00 | 80.00 | 71.00 | 82.00 |
## Usage
To use this model (or others), you can do the following, just changing the model name ([source](https://huggingface.co/dslim/bert-base-NER)):
```
from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
model_name = 'mbeukman/xlm-roberta-base-finetuned-yoruba-finetuned-ner-yoruba'
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
example = "Kò sí ẹ̀rí tí ó fi ẹsẹ̀ rinlẹ̀ ."
ner_results = nlp(example)
print(ner_results)
```
|
mbeukman/xlm-roberta-base-finetuned-yoruba-finetuned-ner-swahili | mbeukman | 2021-11-25T09:05:15Z | 5 | 0 | transformers | [
"transformers",
"pytorch",
"xlm-roberta",
"token-classification",
"NER",
"sw",
"dataset:masakhaner",
"arxiv:2103.11811",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:05Z | ---
language:
- sw
tags:
- NER
datasets:
- masakhaner
metrics:
- f1
- precision
- recall
widget:
- text: "Wizara ya afya ya Tanzania imeripoti Jumatatu kuwa , watu takriban 14 zaidi wamepata maambukizi ya Covid - 19 ."
---
# xlm-roberta-base-finetuned-yoruba-finetuned-ner-swahili
This is a token classification (specifically NER) model that fine-tuned [xlm-roberta-base-finetuned-yoruba](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-yoruba) on the [MasakhaNER](https://arxiv.org/abs/2103.11811) dataset, specifically the Swahili part.
More information, and other similar models can be found in the [main Github repository](https://github.com/Michael-Beukman/NERTransfer).
## About
This model is transformer based and was fine-tuned on the MasakhaNER dataset. It is a named entity recognition dataset, containing mostly news articles in 10 different African languages.
The model was fine-tuned for 50 epochs, with a maximum sequence length of 200, 32 batch size, 5e-5 learning rate. This process was repeated 5 times (with different random seeds), and this uploaded model performed the best out of those 5 seeds (aggregate F1 on test set).
This model was fine-tuned by me, Michael Beukman while doing a project at the University of the Witwatersrand, Johannesburg. This is version 1, as of 20 November 2021.
This model is licensed under the [Apache License, Version 2.0](https://www.apache.org/licenses/LICENSE-2.0).
### Contact & More information
For more information about the models, including training scripts, detailed results and further resources, you can visit the the [main Github repository](https://github.com/Michael-Beukman/NERTransfer). You can contact me by filing an issue on this repository.
### Training Resources
In the interest of openness, and reporting resources used, we list here how long the training process took, as well as what the minimum resources would be to reproduce this. Fine-tuning each model on the NER dataset took between 10 and 30 minutes, and was performed on a NVIDIA RTX3090 GPU. To use a batch size of 32, at least 14GB of GPU memory was required, although it was just possible to fit these models in around 6.5GB's of VRAM when using a batch size of 1.
## Data
The train, evaluation and test datasets were taken directly from the MasakhaNER [Github](https://github.com/masakhane-io/masakhane-ner) repository, with minimal to no preprocessing, as the original dataset is already of high quality.
The motivation for the use of this data is that it is the "first large, publicly available, high quality dataset for named entity recognition (NER) in ten African languages" ([source](https://arxiv.org/pdf/2103.11811.pdf)). The high-quality data, as well as the groundwork laid by the paper introducing it are some more reasons why this dataset was used. For evaluation, the dedicated test split was used, which is from the same distribution as the training data, so this model may not generalise to other distributions, and further testing would need to be done to investigate this. The exact distribution of the data is covered in detail [here](https://arxiv.org/abs/2103.11811).
## Intended Use
This model are intended to be used for NLP research into e.g. interpretability or transfer learning. Using this model in production is not supported, as generalisability and downright performance is limited. In particular, this is not designed to be used in any important downstream task that could affect people, as harm could be caused by the limitations of the model, described next.
## Limitations
This model was only trained on one (relatively small) dataset, covering one task (NER) in one domain (news articles) and in a set span of time. The results may not generalise, and the model may perform badly, or in an unfair / biased way if used on other tasks. Although the purpose of this project was to investigate transfer learning, the performance on languages that the model was not trained for does suffer.
Because this model used xlm-roberta-base as its starting point (potentially with domain adaptive fine-tuning on specific languages), this model's limitations can also apply here. These can include being biased towards the hegemonic viewpoint of most of its training data, being ungrounded and having subpar results on other languages (possibly due to unbalanced training data).
As [Adelani et al. (2021)](https://arxiv.org/abs/2103.11811) showed, the models in general struggled with entities that were either longer than 3 words and entities that were not contained in the training data. This could bias the models towards not finding, e.g. names of people that have many words, possibly leading to a misrepresentation in the results. Similarly, names that are uncommon, and may not have been found in the training data (due to e.g. different languages) would also be predicted less often.
Additionally, this model has not been verified in practice, and other, more subtle problems may become prevalent if used without any verification that it does what it is supposed to.
### Privacy & Ethical Considerations
The data comes from only publicly available news sources, the only available data should cover public figures and those that agreed to be reported on. See the original MasakhaNER paper for more details.
No explicit ethical considerations or adjustments were made during fine-tuning of this model.
## Metrics
The language adaptive models achieve (mostly) superior performance over starting with xlm-roberta-base. Our main metric was the aggregate F1 score for all NER categories.
These metrics are on the test set for MasakhaNER, so the data distribution is similar to the training set, so these results do not directly indicate how well these models generalise.
We do find large variation in transfer results when starting from different seeds (5 different seeds were tested), indicating that the fine-tuning process for transfer might be unstable.
The metrics used were chosen to be consistent with previous work, and to facilitate research. Other metrics may be more appropriate for other purposes.
## Caveats and Recommendations
In general, this model performed worse on the 'date' category compared to others, so if dates are a critical factor, then that might need to be taken into account and addressed, by for example collecting and annotating more data.
## Model Structure
Here are some performance details on this specific model, compared to others we trained.
All of these metrics were calculated on the test set, and the seed was chosen that gave the best overall F1 score. The first three result columns are averaged over all categories, and the latter 4 provide performance broken down by category.
This model can predict the following label for a token ([source](https://huggingface.co/Davlan/xlm-roberta-large-masakhaner)):
Abbreviation|Description
-|-
O|Outside of a named entity
B-DATE |Beginning of a DATE entity right after another DATE entity
I-DATE |DATE entity
B-PER |Beginning of a person’s name right after another person’s name
I-PER |Person’s name
B-ORG |Beginning of an organisation right after another organisation
I-ORG |Organisation
B-LOC |Beginning of a location right after another location
I-LOC |Location
| Model Name | Staring point | Evaluation / Fine-tune Language | F1 | Precision | Recall | F1 (DATE) | F1 (LOC) | F1 (ORG) | F1 (PER) |
| -------------------------------------------------- | -------------------- | -------------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- |
| [xlm-roberta-base-finetuned-yoruba-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-yoruba-finetuned-ner-swahili) (This model) | [yor](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-yoruba) | swa | 87.73 | 86.67 | 88.80 | 85.00 | 91.00 | 75.00 | 93.00 |
| [xlm-roberta-base-finetuned-hausa-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-hausa-finetuned-ner-swahili) | [hau](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-hausa) | swa | 88.36 | 86.95 | 89.82 | 86.00 | 91.00 | 77.00 | 94.00 |
| [xlm-roberta-base-finetuned-igbo-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-igbo-finetuned-ner-swahili) | [ibo](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-igbo) | swa | 87.75 | 86.55 | 88.97 | 85.00 | 92.00 | 77.00 | 91.00 |
| [xlm-roberta-base-finetuned-kinyarwanda-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-kinyarwanda-finetuned-ner-swahili) | [kin](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-kinyarwanda) | swa | 87.26 | 85.15 | 89.48 | 83.00 | 91.00 | 75.00 | 93.00 |
| [xlm-roberta-base-finetuned-luganda-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-luganda-finetuned-ner-swahili) | [lug](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-luganda) | swa | 88.93 | 87.64 | 90.25 | 83.00 | 92.00 | 79.00 | 95.00 |
| [xlm-roberta-base-finetuned-luo-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-luo-finetuned-ner-swahili) | [luo](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-luo) | swa | 87.93 | 86.91 | 88.97 | 83.00 | 91.00 | 76.00 | 94.00 |
| [xlm-roberta-base-finetuned-naija-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-naija-finetuned-ner-swahili) | [pcm](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-naija) | swa | 87.26 | 85.15 | 89.48 | 83.00 | 91.00 | 75.00 | 93.00 |
| [xlm-roberta-base-finetuned-swahili-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-swahili) | [swa](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-swahili) | swa | 90.36 | 88.59 | 92.20 | 86.00 | 93.00 | 79.00 | 96.00 |
| [xlm-roberta-base-finetuned-wolof-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-wolof-finetuned-ner-swahili) | [wol](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-wolof) | swa | 87.80 | 86.50 | 89.14 | 86.00 | 90.00 | 78.00 | 93.00 |
| [xlm-roberta-base-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-ner-swahili) | [base](https://huggingface.co/xlm-roberta-base) | swa | 88.71 | 86.84 | 90.67 | 83.00 | 91.00 | 79.00 | 95.00 |
## Usage
To use this model (or others), you can do the following, just changing the model name ([source](https://huggingface.co/dslim/bert-base-NER)):
```
from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
model_name = 'mbeukman/xlm-roberta-base-finetuned-yoruba-finetuned-ner-swahili'
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
example = "Wizara ya afya ya Tanzania imeripoti Jumatatu kuwa , watu takriban 14 zaidi wamepata maambukizi ya Covid - 19 ."
ner_results = nlp(example)
print(ner_results)
```
|
mbeukman/xlm-roberta-base-finetuned-wolof-finetuned-ner-wolof | mbeukman | 2021-11-25T09:05:13Z | 4 | 0 | transformers | [
"transformers",
"pytorch",
"xlm-roberta",
"token-classification",
"NER",
"wo",
"dataset:masakhaner",
"arxiv:2103.11811",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:05Z | ---
language:
- wo
tags:
- NER
datasets:
- masakhaner
metrics:
- f1
- precision
- recall
widget:
- text: "SAFIYETU BÉEY Céy Koronaa !"
---
# xlm-roberta-base-finetuned-wolof-finetuned-ner-wolof
This is a token classification (specifically NER) model that fine-tuned [xlm-roberta-base-finetuned-wolof](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-wolof) on the [MasakhaNER](https://arxiv.org/abs/2103.11811) dataset, specifically the Wolof part.
More information, and other similar models can be found in the [main Github repository](https://github.com/Michael-Beukman/NERTransfer).
## About
This model is transformer based and was fine-tuned on the MasakhaNER dataset. It is a named entity recognition dataset, containing mostly news articles in 10 different African languages.
The model was fine-tuned for 50 epochs, with a maximum sequence length of 200, 32 batch size, 5e-5 learning rate. This process was repeated 5 times (with different random seeds), and this uploaded model performed the best out of those 5 seeds (aggregate F1 on test set).
This model was fine-tuned by me, Michael Beukman while doing a project at the University of the Witwatersrand, Johannesburg. This is version 1, as of 20 November 2021.
This model is licensed under the [Apache License, Version 2.0](https://www.apache.org/licenses/LICENSE-2.0).
### Contact & More information
For more information about the models, including training scripts, detailed results and further resources, you can visit the the [main Github repository](https://github.com/Michael-Beukman/NERTransfer). You can contact me by filing an issue on this repository.
### Training Resources
In the interest of openness, and reporting resources used, we list here how long the training process took, as well as what the minimum resources would be to reproduce this. Fine-tuning each model on the NER dataset took between 10 and 30 minutes, and was performed on a NVIDIA RTX3090 GPU. To use a batch size of 32, at least 14GB of GPU memory was required, although it was just possible to fit these models in around 6.5GB's of VRAM when using a batch size of 1.
## Data
The train, evaluation and test datasets were taken directly from the MasakhaNER [Github](https://github.com/masakhane-io/masakhane-ner) repository, with minimal to no preprocessing, as the original dataset is already of high quality.
The motivation for the use of this data is that it is the "first large, publicly available, high quality dataset for named entity recognition (NER) in ten African languages" ([source](https://arxiv.org/pdf/2103.11811.pdf)). The high-quality data, as well as the groundwork laid by the paper introducing it are some more reasons why this dataset was used. For evaluation, the dedicated test split was used, which is from the same distribution as the training data, so this model may not generalise to other distributions, and further testing would need to be done to investigate this. The exact distribution of the data is covered in detail [here](https://arxiv.org/abs/2103.11811).
## Intended Use
This model are intended to be used for NLP research into e.g. interpretability or transfer learning. Using this model in production is not supported, as generalisability and downright performance is limited. In particular, this is not designed to be used in any important downstream task that could affect people, as harm could be caused by the limitations of the model, described next.
## Limitations
This model was only trained on one (relatively small) dataset, covering one task (NER) in one domain (news articles) and in a set span of time. The results may not generalise, and the model may perform badly, or in an unfair / biased way if used on other tasks. Although the purpose of this project was to investigate transfer learning, the performance on languages that the model was not trained for does suffer.
Because this model used xlm-roberta-base as its starting point (potentially with domain adaptive fine-tuning on specific languages), this model's limitations can also apply here. These can include being biased towards the hegemonic viewpoint of most of its training data, being ungrounded and having subpar results on other languages (possibly due to unbalanced training data).
As [Adelani et al. (2021)](https://arxiv.org/abs/2103.11811) showed, the models in general struggled with entities that were either longer than 3 words and entities that were not contained in the training data. This could bias the models towards not finding, e.g. names of people that have many words, possibly leading to a misrepresentation in the results. Similarly, names that are uncommon, and may not have been found in the training data (due to e.g. different languages) would also be predicted less often.
Additionally, this model has not been verified in practice, and other, more subtle problems may become prevalent if used without any verification that it does what it is supposed to.
### Privacy & Ethical Considerations
The data comes from only publicly available news sources, the only available data should cover public figures and those that agreed to be reported on. See the original MasakhaNER paper for more details.
No explicit ethical considerations or adjustments were made during fine-tuning of this model.
## Metrics
The language adaptive models achieve (mostly) superior performance over starting with xlm-roberta-base. Our main metric was the aggregate F1 score for all NER categories.
These metrics are on the test set for MasakhaNER, so the data distribution is similar to the training set, so these results do not directly indicate how well these models generalise.
We do find large variation in transfer results when starting from different seeds (5 different seeds were tested), indicating that the fine-tuning process for transfer might be unstable.
The metrics used were chosen to be consistent with previous work, and to facilitate research. Other metrics may be more appropriate for other purposes.
## Caveats and Recommendations
In general, this model performed worse on the 'date' category compared to others, so if dates are a critical factor, then that might need to be taken into account and addressed, by for example collecting and annotating more data.
## Model Structure
Here are some performance details on this specific model, compared to others we trained.
All of these metrics were calculated on the test set, and the seed was chosen that gave the best overall F1 score. The first three result columns are averaged over all categories, and the latter 4 provide performance broken down by category.
This model can predict the following label for a token ([source](https://huggingface.co/Davlan/xlm-roberta-large-masakhaner)):
Abbreviation|Description
-|-
O|Outside of a named entity
B-DATE |Beginning of a DATE entity right after another DATE entity
I-DATE |DATE entity
B-PER |Beginning of a person’s name right after another person’s name
I-PER |Person’s name
B-ORG |Beginning of an organisation right after another organisation
I-ORG |Organisation
B-LOC |Beginning of a location right after another location
I-LOC |Location
| Model Name | Staring point | Evaluation / Fine-tune Language | F1 | Precision | Recall | F1 (DATE) | F1 (LOC) | F1 (ORG) | F1 (PER) |
| -------------------------------------------------- | -------------------- | -------------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- |
| [xlm-roberta-base-finetuned-wolof-finetuned-ner-wolof](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-wolof-finetuned-ner-wolof) (This model) | [wol](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-wolof) | wol | 69.02 | 67.60 | 70.51 | 30.00 | 84.00 | 44.00 | 71.00 |
| [xlm-roberta-base-finetuned-swahili-finetuned-ner-wolof](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-wolof) | [swa](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-swahili) | wol | 69.01 | 73.25 | 65.23 | 27.00 | 85.00 | 52.00 | 67.00 |
| [xlm-roberta-base-finetuned-ner-wolof](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-ner-wolof) | [base](https://huggingface.co/xlm-roberta-base) | wol | 66.12 | 69.46 | 63.09 | 30.00 | 84.00 | 54.00 | 59.00 |
## Usage
To use this model (or others), you can do the following, just changing the model name ([source](https://huggingface.co/dslim/bert-base-NER)):
```
from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
model_name = 'mbeukman/xlm-roberta-base-finetuned-wolof-finetuned-ner-wolof'
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
example = "SAFIYETU BÉEY Céy Koronaa !"
ner_results = nlp(example)
print(ner_results)
```
|
mbeukman/xlm-roberta-base-finetuned-wolof-finetuned-ner-swahili | mbeukman | 2021-11-25T09:05:10Z | 10 | 0 | transformers | [
"transformers",
"pytorch",
"xlm-roberta",
"token-classification",
"NER",
"sw",
"dataset:masakhaner",
"arxiv:2103.11811",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:05Z | ---
language:
- sw
tags:
- NER
datasets:
- masakhaner
metrics:
- f1
- precision
- recall
widget:
- text: "Wizara ya afya ya Tanzania imeripoti Jumatatu kuwa , watu takriban 14 zaidi wamepata maambukizi ya Covid - 19 ."
---
# xlm-roberta-base-finetuned-wolof-finetuned-ner-swahili
This is a token classification (specifically NER) model that fine-tuned [xlm-roberta-base-finetuned-wolof](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-wolof) on the [MasakhaNER](https://arxiv.org/abs/2103.11811) dataset, specifically the Swahili part.
More information, and other similar models can be found in the [main Github repository](https://github.com/Michael-Beukman/NERTransfer).
## About
This model is transformer based and was fine-tuned on the MasakhaNER dataset. It is a named entity recognition dataset, containing mostly news articles in 10 different African languages.
The model was fine-tuned for 50 epochs, with a maximum sequence length of 200, 32 batch size, 5e-5 learning rate. This process was repeated 5 times (with different random seeds), and this uploaded model performed the best out of those 5 seeds (aggregate F1 on test set).
This model was fine-tuned by me, Michael Beukman while doing a project at the University of the Witwatersrand, Johannesburg. This is version 1, as of 20 November 2021.
This model is licensed under the [Apache License, Version 2.0](https://www.apache.org/licenses/LICENSE-2.0).
### Contact & More information
For more information about the models, including training scripts, detailed results and further resources, you can visit the the [main Github repository](https://github.com/Michael-Beukman/NERTransfer). You can contact me by filing an issue on this repository.
### Training Resources
In the interest of openness, and reporting resources used, we list here how long the training process took, as well as what the minimum resources would be to reproduce this. Fine-tuning each model on the NER dataset took between 10 and 30 minutes, and was performed on a NVIDIA RTX3090 GPU. To use a batch size of 32, at least 14GB of GPU memory was required, although it was just possible to fit these models in around 6.5GB's of VRAM when using a batch size of 1.
## Data
The train, evaluation and test datasets were taken directly from the MasakhaNER [Github](https://github.com/masakhane-io/masakhane-ner) repository, with minimal to no preprocessing, as the original dataset is already of high quality.
The motivation for the use of this data is that it is the "first large, publicly available, high quality dataset for named entity recognition (NER) in ten African languages" ([source](https://arxiv.org/pdf/2103.11811.pdf)). The high-quality data, as well as the groundwork laid by the paper introducing it are some more reasons why this dataset was used. For evaluation, the dedicated test split was used, which is from the same distribution as the training data, so this model may not generalise to other distributions, and further testing would need to be done to investigate this. The exact distribution of the data is covered in detail [here](https://arxiv.org/abs/2103.11811).
## Intended Use
This model are intended to be used for NLP research into e.g. interpretability or transfer learning. Using this model in production is not supported, as generalisability and downright performance is limited. In particular, this is not designed to be used in any important downstream task that could affect people, as harm could be caused by the limitations of the model, described next.
## Limitations
This model was only trained on one (relatively small) dataset, covering one task (NER) in one domain (news articles) and in a set span of time. The results may not generalise, and the model may perform badly, or in an unfair / biased way if used on other tasks. Although the purpose of this project was to investigate transfer learning, the performance on languages that the model was not trained for does suffer.
Because this model used xlm-roberta-base as its starting point (potentially with domain adaptive fine-tuning on specific languages), this model's limitations can also apply here. These can include being biased towards the hegemonic viewpoint of most of its training data, being ungrounded and having subpar results on other languages (possibly due to unbalanced training data).
As [Adelani et al. (2021)](https://arxiv.org/abs/2103.11811) showed, the models in general struggled with entities that were either longer than 3 words and entities that were not contained in the training data. This could bias the models towards not finding, e.g. names of people that have many words, possibly leading to a misrepresentation in the results. Similarly, names that are uncommon, and may not have been found in the training data (due to e.g. different languages) would also be predicted less often.
Additionally, this model has not been verified in practice, and other, more subtle problems may become prevalent if used without any verification that it does what it is supposed to.
### Privacy & Ethical Considerations
The data comes from only publicly available news sources, the only available data should cover public figures and those that agreed to be reported on. See the original MasakhaNER paper for more details.
No explicit ethical considerations or adjustments were made during fine-tuning of this model.
## Metrics
The language adaptive models achieve (mostly) superior performance over starting with xlm-roberta-base. Our main metric was the aggregate F1 score for all NER categories.
These metrics are on the test set for MasakhaNER, so the data distribution is similar to the training set, so these results do not directly indicate how well these models generalise.
We do find large variation in transfer results when starting from different seeds (5 different seeds were tested), indicating that the fine-tuning process for transfer might be unstable.
The metrics used were chosen to be consistent with previous work, and to facilitate research. Other metrics may be more appropriate for other purposes.
## Caveats and Recommendations
In general, this model performed worse on the 'date' category compared to others, so if dates are a critical factor, then that might need to be taken into account and addressed, by for example collecting and annotating more data.
## Model Structure
Here are some performance details on this specific model, compared to others we trained.
All of these metrics were calculated on the test set, and the seed was chosen that gave the best overall F1 score. The first three result columns are averaged over all categories, and the latter 4 provide performance broken down by category.
This model can predict the following label for a token ([source](https://huggingface.co/Davlan/xlm-roberta-large-masakhaner)):
Abbreviation|Description
-|-
O|Outside of a named entity
B-DATE |Beginning of a DATE entity right after another DATE entity
I-DATE |DATE entity
B-PER |Beginning of a person’s name right after another person’s name
I-PER |Person’s name
B-ORG |Beginning of an organisation right after another organisation
I-ORG |Organisation
B-LOC |Beginning of a location right after another location
I-LOC |Location
| Model Name | Staring point | Evaluation / Fine-tune Language | F1 | Precision | Recall | F1 (DATE) | F1 (LOC) | F1 (ORG) | F1 (PER) |
| -------------------------------------------------- | -------------------- | -------------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- |
| [xlm-roberta-base-finetuned-wolof-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-wolof-finetuned-ner-swahili) (This model) | [wol](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-wolof) | swa | 87.80 | 86.50 | 89.14 | 86.00 | 90.00 | 78.00 | 93.00 |
| [xlm-roberta-base-finetuned-hausa-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-hausa-finetuned-ner-swahili) | [hau](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-hausa) | swa | 88.36 | 86.95 | 89.82 | 86.00 | 91.00 | 77.00 | 94.00 |
| [xlm-roberta-base-finetuned-igbo-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-igbo-finetuned-ner-swahili) | [ibo](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-igbo) | swa | 87.75 | 86.55 | 88.97 | 85.00 | 92.00 | 77.00 | 91.00 |
| [xlm-roberta-base-finetuned-kinyarwanda-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-kinyarwanda-finetuned-ner-swahili) | [kin](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-kinyarwanda) | swa | 87.26 | 85.15 | 89.48 | 83.00 | 91.00 | 75.00 | 93.00 |
| [xlm-roberta-base-finetuned-luganda-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-luganda-finetuned-ner-swahili) | [lug](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-luganda) | swa | 88.93 | 87.64 | 90.25 | 83.00 | 92.00 | 79.00 | 95.00 |
| [xlm-roberta-base-finetuned-luo-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-luo-finetuned-ner-swahili) | [luo](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-luo) | swa | 87.93 | 86.91 | 88.97 | 83.00 | 91.00 | 76.00 | 94.00 |
| [xlm-roberta-base-finetuned-naija-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-naija-finetuned-ner-swahili) | [pcm](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-naija) | swa | 87.26 | 85.15 | 89.48 | 83.00 | 91.00 | 75.00 | 93.00 |
| [xlm-roberta-base-finetuned-swahili-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-swahili) | [swa](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-swahili) | swa | 90.36 | 88.59 | 92.20 | 86.00 | 93.00 | 79.00 | 96.00 |
| [xlm-roberta-base-finetuned-yoruba-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-yoruba-finetuned-ner-swahili) | [yor](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-yoruba) | swa | 87.73 | 86.67 | 88.80 | 85.00 | 91.00 | 75.00 | 93.00 |
| [xlm-roberta-base-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-ner-swahili) | [base](https://huggingface.co/xlm-roberta-base) | swa | 88.71 | 86.84 | 90.67 | 83.00 | 91.00 | 79.00 | 95.00 |
## Usage
To use this model (or others), you can do the following, just changing the model name ([source](https://huggingface.co/dslim/bert-base-NER)):
```
from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
model_name = 'mbeukman/xlm-roberta-base-finetuned-wolof-finetuned-ner-swahili'
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
example = "Wizara ya afya ya Tanzania imeripoti Jumatatu kuwa , watu takriban 14 zaidi wamepata maambukizi ya Covid - 19 ."
ner_results = nlp(example)
print(ner_results)
```
|
mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-wolof | mbeukman | 2021-11-25T09:05:05Z | 4 | 0 | transformers | [
"transformers",
"pytorch",
"xlm-roberta",
"token-classification",
"NER",
"wo",
"dataset:masakhaner",
"arxiv:2103.11811",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:05Z | ---
language:
- wo
tags:
- NER
datasets:
- masakhaner
metrics:
- f1
- precision
- recall
widget:
- text: "SAFIYETU BÉEY Céy Koronaa !"
---
# xlm-roberta-base-finetuned-swahili-finetuned-ner-wolof
This is a token classification (specifically NER) model that fine-tuned [xlm-roberta-base-finetuned-swahili](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-swahili) on the [MasakhaNER](https://arxiv.org/abs/2103.11811) dataset, specifically the Wolof part.
More information, and other similar models can be found in the [main Github repository](https://github.com/Michael-Beukman/NERTransfer).
## About
This model is transformer based and was fine-tuned on the MasakhaNER dataset. It is a named entity recognition dataset, containing mostly news articles in 10 different African languages.
The model was fine-tuned for 50 epochs, with a maximum sequence length of 200, 32 batch size, 5e-5 learning rate. This process was repeated 5 times (with different random seeds), and this uploaded model performed the best out of those 5 seeds (aggregate F1 on test set).
This model was fine-tuned by me, Michael Beukman while doing a project at the University of the Witwatersrand, Johannesburg. This is version 1, as of 20 November 2021.
This model is licensed under the [Apache License, Version 2.0](https://www.apache.org/licenses/LICENSE-2.0).
### Contact & More information
For more information about the models, including training scripts, detailed results and further resources, you can visit the the [main Github repository](https://github.com/Michael-Beukman/NERTransfer). You can contact me by filing an issue on this repository.
### Training Resources
In the interest of openness, and reporting resources used, we list here how long the training process took, as well as what the minimum resources would be to reproduce this. Fine-tuning each model on the NER dataset took between 10 and 30 minutes, and was performed on a NVIDIA RTX3090 GPU. To use a batch size of 32, at least 14GB of GPU memory was required, although it was just possible to fit these models in around 6.5GB's of VRAM when using a batch size of 1.
## Data
The train, evaluation and test datasets were taken directly from the MasakhaNER [Github](https://github.com/masakhane-io/masakhane-ner) repository, with minimal to no preprocessing, as the original dataset is already of high quality.
The motivation for the use of this data is that it is the "first large, publicly available, high quality dataset for named entity recognition (NER) in ten African languages" ([source](https://arxiv.org/pdf/2103.11811.pdf)). The high-quality data, as well as the groundwork laid by the paper introducing it are some more reasons why this dataset was used. For evaluation, the dedicated test split was used, which is from the same distribution as the training data, so this model may not generalise to other distributions, and further testing would need to be done to investigate this. The exact distribution of the data is covered in detail [here](https://arxiv.org/abs/2103.11811).
## Intended Use
This model are intended to be used for NLP research into e.g. interpretability or transfer learning. Using this model in production is not supported, as generalisability and downright performance is limited. In particular, this is not designed to be used in any important downstream task that could affect people, as harm could be caused by the limitations of the model, described next.
## Limitations
This model was only trained on one (relatively small) dataset, covering one task (NER) in one domain (news articles) and in a set span of time. The results may not generalise, and the model may perform badly, or in an unfair / biased way if used on other tasks. Although the purpose of this project was to investigate transfer learning, the performance on languages that the model was not trained for does suffer.
Because this model used xlm-roberta-base as its starting point (potentially with domain adaptive fine-tuning on specific languages), this model's limitations can also apply here. These can include being biased towards the hegemonic viewpoint of most of its training data, being ungrounded and having subpar results on other languages (possibly due to unbalanced training data).
As [Adelani et al. (2021)](https://arxiv.org/abs/2103.11811) showed, the models in general struggled with entities that were either longer than 3 words and entities that were not contained in the training data. This could bias the models towards not finding, e.g. names of people that have many words, possibly leading to a misrepresentation in the results. Similarly, names that are uncommon, and may not have been found in the training data (due to e.g. different languages) would also be predicted less often.
Additionally, this model has not been verified in practice, and other, more subtle problems may become prevalent if used without any verification that it does what it is supposed to.
### Privacy & Ethical Considerations
The data comes from only publicly available news sources, the only available data should cover public figures and those that agreed to be reported on. See the original MasakhaNER paper for more details.
No explicit ethical considerations or adjustments were made during fine-tuning of this model.
## Metrics
The language adaptive models achieve (mostly) superior performance over starting with xlm-roberta-base. Our main metric was the aggregate F1 score for all NER categories.
These metrics are on the test set for MasakhaNER, so the data distribution is similar to the training set, so these results do not directly indicate how well these models generalise.
We do find large variation in transfer results when starting from different seeds (5 different seeds were tested), indicating that the fine-tuning process for transfer might be unstable.
The metrics used were chosen to be consistent with previous work, and to facilitate research. Other metrics may be more appropriate for other purposes.
## Caveats and Recommendations
In general, this model performed worse on the 'date' category compared to others, so if dates are a critical factor, then that might need to be taken into account and addressed, by for example collecting and annotating more data.
## Model Structure
Here are some performance details on this specific model, compared to others we trained.
All of these metrics were calculated on the test set, and the seed was chosen that gave the best overall F1 score. The first three result columns are averaged over all categories, and the latter 4 provide performance broken down by category.
This model can predict the following label for a token ([source](https://huggingface.co/Davlan/xlm-roberta-large-masakhaner)):
Abbreviation|Description
-|-
O|Outside of a named entity
B-DATE |Beginning of a DATE entity right after another DATE entity
I-DATE |DATE entity
B-PER |Beginning of a person’s name right after another person’s name
I-PER |Person’s name
B-ORG |Beginning of an organisation right after another organisation
I-ORG |Organisation
B-LOC |Beginning of a location right after another location
I-LOC |Location
| Model Name | Staring point | Evaluation / Fine-tune Language | F1 | Precision | Recall | F1 (DATE) | F1 (LOC) | F1 (ORG) | F1 (PER) |
| -------------------------------------------------- | -------------------- | -------------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- |
| [xlm-roberta-base-finetuned-swahili-finetuned-ner-wolof](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-wolof) (This model) | [swa](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-swahili) | wol | 69.01 | 73.25 | 65.23 | 27.00 | 85.00 | 52.00 | 67.00 |
| [xlm-roberta-base-finetuned-wolof-finetuned-ner-wolof](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-wolof-finetuned-ner-wolof) | [wol](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-wolof) | wol | 69.02 | 67.60 | 70.51 | 30.00 | 84.00 | 44.00 | 71.00 |
| [xlm-roberta-base-finetuned-ner-wolof](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-ner-wolof) | [base](https://huggingface.co/xlm-roberta-base) | wol | 66.12 | 69.46 | 63.09 | 30.00 | 84.00 | 54.00 | 59.00 |
## Usage
To use this model (or others), you can do the following, just changing the model name ([source](https://huggingface.co/dslim/bert-base-NER)):
```
from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
model_name = 'mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-wolof'
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
example = "SAFIYETU BÉEY Céy Koronaa !"
ner_results = nlp(example)
print(ner_results)
```
|
mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-luo | mbeukman | 2021-11-25T09:04:58Z | 14 | 0 | transformers | [
"transformers",
"pytorch",
"xlm-roberta",
"token-classification",
"NER",
"luo",
"dataset:masakhaner",
"arxiv:2103.11811",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:05Z | ---
language:
- luo
tags:
- NER
datasets:
- masakhaner
metrics:
- f1
- precision
- recall
widget:
- text: "Jii 2 moko jowito ngimagi ka machielo 1 to ohinyore marach mokalo e masira makoch mar apaya mane otimore e apaya mawuok Oyugis kochimo Chabera e sub county ma Rachuonyo East e County ma Homa Bay ewii odhiambo makawuononi"
---
# xlm-roberta-base-finetuned-swahili-finetuned-ner-luo
This is a token classification (specifically NER) model that fine-tuned [xlm-roberta-base-finetuned-swahili](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-swahili) on the [MasakhaNER](https://arxiv.org/abs/2103.11811) dataset, specifically the Luo part.
More information, and other similar models can be found in the [main Github repository](https://github.com/Michael-Beukman/NERTransfer).
## About
This model is transformer based and was fine-tuned on the MasakhaNER dataset. It is a named entity recognition dataset, containing mostly news articles in 10 different African languages.
The model was fine-tuned for 50 epochs, with a maximum sequence length of 200, 32 batch size, 5e-5 learning rate. This process was repeated 5 times (with different random seeds), and this uploaded model performed the best out of those 5 seeds (aggregate F1 on test set).
This model was fine-tuned by me, Michael Beukman while doing a project at the University of the Witwatersrand, Johannesburg. This is version 1, as of 20 November 2021.
This model is licensed under the [Apache License, Version 2.0](https://www.apache.org/licenses/LICENSE-2.0).
### Contact & More information
For more information about the models, including training scripts, detailed results and further resources, you can visit the the [main Github repository](https://github.com/Michael-Beukman/NERTransfer). You can contact me by filing an issue on this repository.
### Training Resources
In the interest of openness, and reporting resources used, we list here how long the training process took, as well as what the minimum resources would be to reproduce this. Fine-tuning each model on the NER dataset took between 10 and 30 minutes, and was performed on a NVIDIA RTX3090 GPU. To use a batch size of 32, at least 14GB of GPU memory was required, although it was just possible to fit these models in around 6.5GB's of VRAM when using a batch size of 1.
## Data
The train, evaluation and test datasets were taken directly from the MasakhaNER [Github](https://github.com/masakhane-io/masakhane-ner) repository, with minimal to no preprocessing, as the original dataset is already of high quality.
The motivation for the use of this data is that it is the "first large, publicly available, high quality dataset for named entity recognition (NER) in ten African languages" ([source](https://arxiv.org/pdf/2103.11811.pdf)). The high-quality data, as well as the groundwork laid by the paper introducing it are some more reasons why this dataset was used. For evaluation, the dedicated test split was used, which is from the same distribution as the training data, so this model may not generalise to other distributions, and further testing would need to be done to investigate this. The exact distribution of the data is covered in detail [here](https://arxiv.org/abs/2103.11811).
## Intended Use
This model are intended to be used for NLP research into e.g. interpretability or transfer learning. Using this model in production is not supported, as generalisability and downright performance is limited. In particular, this is not designed to be used in any important downstream task that could affect people, as harm could be caused by the limitations of the model, described next.
## Limitations
This model was only trained on one (relatively small) dataset, covering one task (NER) in one domain (news articles) and in a set span of time. The results may not generalise, and the model may perform badly, or in an unfair / biased way if used on other tasks. Although the purpose of this project was to investigate transfer learning, the performance on languages that the model was not trained for does suffer.
Because this model used xlm-roberta-base as its starting point (potentially with domain adaptive fine-tuning on specific languages), this model's limitations can also apply here. These can include being biased towards the hegemonic viewpoint of most of its training data, being ungrounded and having subpar results on other languages (possibly due to unbalanced training data).
As [Adelani et al. (2021)](https://arxiv.org/abs/2103.11811) showed, the models in general struggled with entities that were either longer than 3 words and entities that were not contained in the training data. This could bias the models towards not finding, e.g. names of people that have many words, possibly leading to a misrepresentation in the results. Similarly, names that are uncommon, and may not have been found in the training data (due to e.g. different languages) would also be predicted less often.
Additionally, this model has not been verified in practice, and other, more subtle problems may become prevalent if used without any verification that it does what it is supposed to.
### Privacy & Ethical Considerations
The data comes from only publicly available news sources, the only available data should cover public figures and those that agreed to be reported on. See the original MasakhaNER paper for more details.
No explicit ethical considerations or adjustments were made during fine-tuning of this model.
## Metrics
The language adaptive models achieve (mostly) superior performance over starting with xlm-roberta-base. Our main metric was the aggregate F1 score for all NER categories.
These metrics are on the test set for MasakhaNER, so the data distribution is similar to the training set, so these results do not directly indicate how well these models generalise.
We do find large variation in transfer results when starting from different seeds (5 different seeds were tested), indicating that the fine-tuning process for transfer might be unstable.
The metrics used were chosen to be consistent with previous work, and to facilitate research. Other metrics may be more appropriate for other purposes.
## Caveats and Recommendations
In general, this model performed worse on the 'date' category compared to others, so if dates are a critical factor, then that might need to be taken into account and addressed, by for example collecting and annotating more data.
## Model Structure
Here are some performance details on this specific model, compared to others we trained.
All of these metrics were calculated on the test set, and the seed was chosen that gave the best overall F1 score. The first three result columns are averaged over all categories, and the latter 4 provide performance broken down by category.
This model can predict the following label for a token ([source](https://huggingface.co/Davlan/xlm-roberta-large-masakhaner)):
Abbreviation|Description
-|-
O|Outside of a named entity
B-DATE |Beginning of a DATE entity right after another DATE entity
I-DATE |DATE entity
B-PER |Beginning of a person’s name right after another person’s name
I-PER |Person’s name
B-ORG |Beginning of an organisation right after another organisation
I-ORG |Organisation
B-LOC |Beginning of a location right after another location
I-LOC |Location
| Model Name | Staring point | Evaluation / Fine-tune Language | F1 | Precision | Recall | F1 (DATE) | F1 (LOC) | F1 (ORG) | F1 (PER) |
| -------------------------------------------------- | -------------------- | -------------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- |
| [xlm-roberta-base-finetuned-swahili-finetuned-ner-luo](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-luo) (This model) | [swa](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-swahili) | luo | 78.13 | 77.75 | 78.52 | 65.00 | 82.00 | 61.00 | 89.00 |
| [xlm-roberta-base-finetuned-luo-finetuned-ner-luo](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-luo-finetuned-ner-luo) | [luo](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-luo) | luo | 78.71 | 78.91 | 78.52 | 72.00 | 84.00 | 59.00 | 87.00 |
| [xlm-roberta-base-finetuned-ner-luo](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-ner-luo) | [base](https://huggingface.co/xlm-roberta-base) | luo | 75.99 | 76.18 | 75.80 | 71.00 | 76.00 | 62.00 | 85.00 |
## Usage
To use this model (or others), you can do the following, just changing the model name ([source](https://huggingface.co/dslim/bert-base-NER)):
```
from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
model_name = 'mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-luo'
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
example = "Jii 2 moko jowito ngimagi ka machielo 1 to ohinyore marach mokalo e masira makoch mar apaya mane otimore e apaya mawuok Oyugis kochimo Chabera e sub county ma Rachuonyo East e County ma Homa Bay ewii odhiambo makawuononi"
ner_results = nlp(example)
print(ner_results)
```
|
mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-luganda | mbeukman | 2021-11-25T09:04:55Z | 3 | 0 | transformers | [
"transformers",
"pytorch",
"xlm-roberta",
"token-classification",
"NER",
"lug",
"dataset:masakhaner",
"arxiv:2103.11811",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:05Z | ---
language:
- lug
tags:
- NER
datasets:
- masakhaner
metrics:
- f1
- precision
- recall
widget:
- text: "Empaka zaakubeera mu kibuga Liverpool e Bungereza , okutandika nga July 12 ."
---
# xlm-roberta-base-finetuned-swahili-finetuned-ner-luganda
This is a token classification (specifically NER) model that fine-tuned [xlm-roberta-base-finetuned-swahili](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-swahili) on the [MasakhaNER](https://arxiv.org/abs/2103.11811) dataset, specifically the luganda part.
More information, and other similar models can be found in the [main Github repository](https://github.com/Michael-Beukman/NERTransfer).
## About
This model is transformer based and was fine-tuned on the MasakhaNER dataset. It is a named entity recognition dataset, containing mostly news articles in 10 different African languages.
The model was fine-tuned for 50 epochs, with a maximum sequence length of 200, 32 batch size, 5e-5 learning rate. This process was repeated 5 times (with different random seeds), and this uploaded model performed the best out of those 5 seeds (aggregate F1 on test set).
This model was fine-tuned by me, Michael Beukman while doing a project at the University of the Witwatersrand, Johannesburg. This is version 1, as of 20 November 2021.
This model is licensed under the [Apache License, Version 2.0](https://www.apache.org/licenses/LICENSE-2.0).
### Contact & More information
For more information about the models, including training scripts, detailed results and further resources, you can visit the the [main Github repository](https://github.com/Michael-Beukman/NERTransfer). You can contact me by filing an issue on this repository.
### Training Resources
In the interest of openness, and reporting resources used, we list here how long the training process took, as well as what the minimum resources would be to reproduce this. Fine-tuning each model on the NER dataset took between 10 and 30 minutes, and was performed on a NVIDIA RTX3090 GPU. To use a batch size of 32, at least 14GB of GPU memory was required, although it was just possible to fit these models in around 6.5GB's of VRAM when using a batch size of 1.
## Data
The train, evaluation and test datasets were taken directly from the MasakhaNER [Github](https://github.com/masakhane-io/masakhane-ner) repository, with minimal to no preprocessing, as the original dataset is already of high quality.
The motivation for the use of this data is that it is the "first large, publicly available, high quality dataset for named entity recognition (NER) in ten African languages" ([source](https://arxiv.org/pdf/2103.11811.pdf)). The high-quality data, as well as the groundwork laid by the paper introducing it are some more reasons why this dataset was used. For evaluation, the dedicated test split was used, which is from the same distribution as the training data, so this model may not generalise to other distributions, and further testing would need to be done to investigate this. The exact distribution of the data is covered in detail [here](https://arxiv.org/abs/2103.11811).
## Intended Use
This model are intended to be used for NLP research into e.g. interpretability or transfer learning. Using this model in production is not supported, as generalisability and downright performance is limited. In particular, this is not designed to be used in any important downstream task that could affect people, as harm could be caused by the limitations of the model, described next.
## Limitations
This model was only trained on one (relatively small) dataset, covering one task (NER) in one domain (news articles) and in a set span of time. The results may not generalise, and the model may perform badly, or in an unfair / biased way if used on other tasks. Although the purpose of this project was to investigate transfer learning, the performance on languages that the model was not trained for does suffer.
Because this model used xlm-roberta-base as its starting point (potentially with domain adaptive fine-tuning on specific languages), this model's limitations can also apply here. These can include being biased towards the hegemonic viewpoint of most of its training data, being ungrounded and having subpar results on other languages (possibly due to unbalanced training data).
As [Adelani et al. (2021)](https://arxiv.org/abs/2103.11811) showed, the models in general struggled with entities that were either longer than 3 words and entities that were not contained in the training data. This could bias the models towards not finding, e.g. names of people that have many words, possibly leading to a misrepresentation in the results. Similarly, names that are uncommon, and may not have been found in the training data (due to e.g. different languages) would also be predicted less often.
Additionally, this model has not been verified in practice, and other, more subtle problems may become prevalent if used without any verification that it does what it is supposed to.
### Privacy & Ethical Considerations
The data comes from only publicly available news sources, the only available data should cover public figures and those that agreed to be reported on. See the original MasakhaNER paper for more details.
No explicit ethical considerations or adjustments were made during fine-tuning of this model.
## Metrics
The language adaptive models achieve (mostly) superior performance over starting with xlm-roberta-base. Our main metric was the aggregate F1 score for all NER categories.
These metrics are on the test set for MasakhaNER, so the data distribution is similar to the training set, so these results do not directly indicate how well these models generalise.
We do find large variation in transfer results when starting from different seeds (5 different seeds were tested), indicating that the fine-tuning process for transfer might be unstable.
The metrics used were chosen to be consistent with previous work, and to facilitate research. Other metrics may be more appropriate for other purposes.
## Caveats and Recommendations
In general, this model performed worse on the 'date' category compared to others, so if dates are a critical factor, then that might need to be taken into account and addressed, by for example collecting and annotating more data.
## Model Structure
Here are some performance details on this specific model, compared to others we trained.
All of these metrics were calculated on the test set, and the seed was chosen that gave the best overall F1 score. The first three result columns are averaged over all categories, and the latter 4 provide performance broken down by category.
This model can predict the following label for a token ([source](https://huggingface.co/Davlan/xlm-roberta-large-masakhaner)):
Abbreviation|Description
-|-
O|Outside of a named entity
B-DATE |Beginning of a DATE entity right after another DATE entity
I-DATE |DATE entity
B-PER |Beginning of a person’s name right after another person’s name
I-PER |Person’s name
B-ORG |Beginning of an organisation right after another organisation
I-ORG |Organisation
B-LOC |Beginning of a location right after another location
I-LOC |Location
| Model Name | Staring point | Evaluation / Fine-tune Language | F1 | Precision | Recall | F1 (DATE) | F1 (LOC) | F1 (ORG) | F1 (PER) |
| -------------------------------------------------- | -------------------- | -------------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- |
| [xlm-roberta-base-finetuned-swahili-finetuned-ner-luganda](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-luganda) (This model) | [swa](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-swahili) | lug | 82.57 | 80.38 | 84.89 | 75.00 | 80.00 | 82.00 | 87.00 |
| [xlm-roberta-base-finetuned-luganda-finetuned-ner-luganda](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-luganda-finetuned-ner-luganda) | [lug](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-luganda) | lug | 85.37 | 82.75 | 88.17 | 78.00 | 82.00 | 80.00 | 92.00 |
| [xlm-roberta-base-finetuned-ner-luganda](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-ner-luganda) | [base](https://huggingface.co/xlm-roberta-base) | lug | 80.91 | 78.59 | 83.37 | 73.00 | 78.00 | 77.00 | 86.00 |
## Usage
To use this model (or others), you can do the following, just changing the model name ([source](https://huggingface.co/dslim/bert-base-NER)):
```
from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
model_name = 'mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-luganda'
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
example = "Empaka zaakubeera mu kibuga Liverpool e Bungereza , okutandika nga July 12 ."
ner_results = nlp(example)
print(ner_results)
```
|
mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-hausa | mbeukman | 2021-11-25T09:04:48Z | 5 | 0 | transformers | [
"transformers",
"pytorch",
"xlm-roberta",
"token-classification",
"NER",
"ha",
"dataset:masakhaner",
"arxiv:2103.11811",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:05Z | ---
language:
- ha
tags:
- NER
datasets:
- masakhaner
metrics:
- f1
- precision
- recall
widget:
- text: "A saurari cikakken rahoton wakilin Muryar Amurka Ibrahim Abdul'aziz"
---
# xlm-roberta-base-finetuned-swahili-finetuned-ner-hausa
This is a token classification (specifically NER) model that fine-tuned [xlm-roberta-base-finetuned-swahili](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-swahili) on the [MasakhaNER](https://arxiv.org/abs/2103.11811) dataset, specifically the Hausa part.
More information, and other similar models can be found in the [main Github repository](https://github.com/Michael-Beukman/NERTransfer).
## About
This model is transformer based and was fine-tuned on the MasakhaNER dataset. It is a named entity recognition dataset, containing mostly news articles in 10 different African languages.
The model was fine-tuned for 50 epochs, with a maximum sequence length of 200, 32 batch size, 5e-5 learning rate. This process was repeated 5 times (with different random seeds), and this uploaded model performed the best out of those 5 seeds (aggregate F1 on test set).
This model was fine-tuned by me, Michael Beukman while doing a project at the University of the Witwatersrand, Johannesburg. This is version 1, as of 20 November 2021.
This model is licensed under the [Apache License, Version 2.0](https://www.apache.org/licenses/LICENSE-2.0).
### Contact & More information
For more information about the models, including training scripts, detailed results and further resources, you can visit the the [main Github repository](https://github.com/Michael-Beukman/NERTransfer). You can contact me by filing an issue on this repository.
### Training Resources
In the interest of openness, and reporting resources used, we list here how long the training process took, as well as what the minimum resources would be to reproduce this. Fine-tuning each model on the NER dataset took between 10 and 30 minutes, and was performed on a NVIDIA RTX3090 GPU. To use a batch size of 32, at least 14GB of GPU memory was required, although it was just possible to fit these models in around 6.5GB's of VRAM when using a batch size of 1.
## Data
The train, evaluation and test datasets were taken directly from the MasakhaNER [Github](https://github.com/masakhane-io/masakhane-ner) repository, with minimal to no preprocessing, as the original dataset is already of high quality.
The motivation for the use of this data is that it is the "first large, publicly available, high quality dataset for named entity recognition (NER) in ten African languages" ([source](https://arxiv.org/pdf/2103.11811.pdf)). The high-quality data, as well as the groundwork laid by the paper introducing it are some more reasons why this dataset was used. For evaluation, the dedicated test split was used, which is from the same distribution as the training data, so this model may not generalise to other distributions, and further testing would need to be done to investigate this. The exact distribution of the data is covered in detail [here](https://arxiv.org/abs/2103.11811).
## Intended Use
This model are intended to be used for NLP research into e.g. interpretability or transfer learning. Using this model in production is not supported, as generalisability and downright performance is limited. In particular, this is not designed to be used in any important downstream task that could affect people, as harm could be caused by the limitations of the model, described next.
## Limitations
This model was only trained on one (relatively small) dataset, covering one task (NER) in one domain (news articles) and in a set span of time. The results may not generalise, and the model may perform badly, or in an unfair / biased way if used on other tasks. Although the purpose of this project was to investigate transfer learning, the performance on languages that the model was not trained for does suffer.
Because this model used xlm-roberta-base as its starting point (potentially with domain adaptive fine-tuning on specific languages), this model's limitations can also apply here. These can include being biased towards the hegemonic viewpoint of most of its training data, being ungrounded and having subpar results on other languages (possibly due to unbalanced training data).
As [Adelani et al. (2021)](https://arxiv.org/abs/2103.11811) showed, the models in general struggled with entities that were either longer than 3 words and entities that were not contained in the training data. This could bias the models towards not finding, e.g. names of people that have many words, possibly leading to a misrepresentation in the results. Similarly, names that are uncommon, and may not have been found in the training data (due to e.g. different languages) would also be predicted less often.
Additionally, this model has not been verified in practice, and other, more subtle problems may become prevalent if used without any verification that it does what it is supposed to.
### Privacy & Ethical Considerations
The data comes from only publicly available news sources, the only available data should cover public figures and those that agreed to be reported on. See the original MasakhaNER paper for more details.
No explicit ethical considerations or adjustments were made during fine-tuning of this model.
## Metrics
The language adaptive models achieve (mostly) superior performance over starting with xlm-roberta-base. Our main metric was the aggregate F1 score for all NER categories.
These metrics are on the test set for MasakhaNER, so the data distribution is similar to the training set, so these results do not directly indicate how well these models generalise.
We do find large variation in transfer results when starting from different seeds (5 different seeds were tested), indicating that the fine-tuning process for transfer might be unstable.
The metrics used were chosen to be consistent with previous work, and to facilitate research. Other metrics may be more appropriate for other purposes.
## Caveats and Recommendations
In general, this model performed worse on the 'date' category compared to others, so if dates are a critical factor, then that might need to be taken into account and addressed, by for example collecting and annotating more data.
## Model Structure
Here are some performance details on this specific model, compared to others we trained.
All of these metrics were calculated on the test set, and the seed was chosen that gave the best overall F1 score. The first three result columns are averaged over all categories, and the latter 4 provide performance broken down by category.
This model can predict the following label for a token ([source](https://huggingface.co/Davlan/xlm-roberta-large-masakhaner)):
Abbreviation|Description
-|-
O|Outside of a named entity
B-DATE |Beginning of a DATE entity right after another DATE entity
I-DATE |DATE entity
B-PER |Beginning of a person’s name right after another person’s name
I-PER |Person’s name
B-ORG |Beginning of an organisation right after another organisation
I-ORG |Organisation
B-LOC |Beginning of a location right after another location
I-LOC |Location
| Model Name | Staring point | Evaluation / Fine-tune Language | F1 | Precision | Recall | F1 (DATE) | F1 (LOC) | F1 (ORG) | F1 (PER) |
| -------------------------------------------------- | -------------------- | -------------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- |
| [xlm-roberta-base-finetuned-swahili-finetuned-ner-hausa](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-hausa) (This model) | [swa](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-swahili) | hau | 89.14 | 87.18 | 91.20 | 82.00 | 93.00 | 76.00 | 93.00 |
| [xlm-roberta-base-finetuned-hausa-finetuned-ner-hausa](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-hausa-finetuned-ner-hausa) | [hau](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-hausa) | hau | 92.27 | 90.46 | 94.16 | 85.00 | 95.00 | 80.00 | 97.00 |
| [xlm-roberta-base-finetuned-ner-hausa](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-ner-hausa) | [base](https://huggingface.co/xlm-roberta-base) | hau | 89.94 | 87.74 | 92.25 | 84.00 | 94.00 | 74.00 | 93.00 |
## Usage
To use this model (or others), you can do the following, just changing the model name ([source](https://huggingface.co/dslim/bert-base-NER)):
```
from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
model_name = 'mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-hausa'
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
example = "A saurari cikakken rahoton wakilin Muryar Amurka Ibrahim Abdul'aziz"
ner_results = nlp(example)
print(ner_results)
```
|
mbeukman/xlm-roberta-base-finetuned-ner-luo | mbeukman | 2021-11-25T09:04:35Z | 7 | 0 | transformers | [
"transformers",
"pytorch",
"xlm-roberta",
"token-classification",
"NER",
"luo",
"dataset:masakhaner",
"arxiv:2103.11811",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:05Z | ---
language:
- luo
tags:
- NER
datasets:
- masakhaner
metrics:
- f1
- precision
- recall
widget:
- text: "Jii 2 moko jowito ngimagi ka machielo 1 to ohinyore marach mokalo e masira makoch mar apaya mane otimore e apaya mawuok Oyugis kochimo Chabera e sub county ma Rachuonyo East e County ma Homa Bay ewii odhiambo makawuononi"
---
# xlm-roberta-base-finetuned-ner-luo
This is a token classification (specifically NER) model that fine-tuned [xlm-roberta-base](https://huggingface.co/xlm-roberta-base) on the [MasakhaNER](https://arxiv.org/abs/2103.11811) dataset, specifically the Luo part.
More information, and other similar models can be found in the [main Github repository](https://github.com/Michael-Beukman/NERTransfer).
## About
This model is transformer based and was fine-tuned on the MasakhaNER dataset. It is a named entity recognition dataset, containing mostly news articles in 10 different African languages.
The model was fine-tuned for 50 epochs, with a maximum sequence length of 200, 32 batch size, 5e-5 learning rate. This process was repeated 5 times (with different random seeds), and this uploaded model performed the best out of those 5 seeds (aggregate F1 on test set).
This model was fine-tuned by me, Michael Beukman while doing a project at the University of the Witwatersrand, Johannesburg. This is version 1, as of 20 November 2021.
This model is licensed under the [Apache License, Version 2.0](https://www.apache.org/licenses/LICENSE-2.0).
### Contact & More information
For more information about the models, including training scripts, detailed results and further resources, you can visit the the [main Github repository](https://github.com/Michael-Beukman/NERTransfer). You can contact me by filing an issue on this repository.
### Training Resources
In the interest of openness, and reporting resources used, we list here how long the training process took, as well as what the minimum resources would be to reproduce this. Fine-tuning each model on the NER dataset took between 10 and 30 minutes, and was performed on a NVIDIA RTX3090 GPU. To use a batch size of 32, at least 14GB of GPU memory was required, although it was just possible to fit these models in around 6.5GB's of VRAM when using a batch size of 1.
## Data
The train, evaluation and test datasets were taken directly from the MasakhaNER [Github](https://github.com/masakhane-io/masakhane-ner) repository, with minimal to no preprocessing, as the original dataset is already of high quality.
The motivation for the use of this data is that it is the "first large, publicly available, high quality dataset for named entity recognition (NER) in ten African languages" ([source](https://arxiv.org/pdf/2103.11811.pdf)). The high-quality data, as well as the groundwork laid by the paper introducing it are some more reasons why this dataset was used. For evaluation, the dedicated test split was used, which is from the same distribution as the training data, so this model may not generalise to other distributions, and further testing would need to be done to investigate this. The exact distribution of the data is covered in detail [here](https://arxiv.org/abs/2103.11811).
## Intended Use
This model are intended to be used for NLP research into e.g. interpretability or transfer learning. Using this model in production is not supported, as generalisability and downright performance is limited. In particular, this is not designed to be used in any important downstream task that could affect people, as harm could be caused by the limitations of the model, described next.
## Limitations
This model was only trained on one (relatively small) dataset, covering one task (NER) in one domain (news articles) and in a set span of time. The results may not generalise, and the model may perform badly, or in an unfair / biased way if used on other tasks. Although the purpose of this project was to investigate transfer learning, the performance on languages that the model was not trained for does suffer.
Because this model used xlm-roberta-base as its starting point (potentially with domain adaptive fine-tuning on specific languages), this model's limitations can also apply here. These can include being biased towards the hegemonic viewpoint of most of its training data, being ungrounded and having subpar results on other languages (possibly due to unbalanced training data).
As [Adelani et al. (2021)](https://arxiv.org/abs/2103.11811) showed, the models in general struggled with entities that were either longer than 3 words and entities that were not contained in the training data. This could bias the models towards not finding, e.g. names of people that have many words, possibly leading to a misrepresentation in the results. Similarly, names that are uncommon, and may not have been found in the training data (due to e.g. different languages) would also be predicted less often.
Additionally, this model has not been verified in practice, and other, more subtle problems may become prevalent if used without any verification that it does what it is supposed to.
### Privacy & Ethical Considerations
The data comes from only publicly available news sources, the only available data should cover public figures and those that agreed to be reported on. See the original MasakhaNER paper for more details.
No explicit ethical considerations or adjustments were made during fine-tuning of this model.
## Metrics
The language adaptive models achieve (mostly) superior performance over starting with xlm-roberta-base. Our main metric was the aggregate F1 score for all NER categories.
These metrics are on the test set for MasakhaNER, so the data distribution is similar to the training set, so these results do not directly indicate how well these models generalise.
We do find large variation in transfer results when starting from different seeds (5 different seeds were tested), indicating that the fine-tuning process for transfer might be unstable.
The metrics used were chosen to be consistent with previous work, and to facilitate research. Other metrics may be more appropriate for other purposes.
## Caveats and Recommendations
In general, this model performed worse on the 'date' category compared to others, so if dates are a critical factor, then that might need to be taken into account and addressed, by for example collecting and annotating more data.
## Model Structure
Here are some performance details on this specific model, compared to others we trained.
All of these metrics were calculated on the test set, and the seed was chosen that gave the best overall F1 score. The first three result columns are averaged over all categories, and the latter 4 provide performance broken down by category.
This model can predict the following label for a token ([source](https://huggingface.co/Davlan/xlm-roberta-large-masakhaner)):
Abbreviation|Description
-|-
O|Outside of a named entity
B-DATE |Beginning of a DATE entity right after another DATE entity
I-DATE |DATE entity
B-PER |Beginning of a person’s name right after another person’s name
I-PER |Person’s name
B-ORG |Beginning of an organisation right after another organisation
I-ORG |Organisation
B-LOC |Beginning of a location right after another location
I-LOC |Location
| Model Name | Staring point | Evaluation / Fine-tune Language | F1 | Precision | Recall | F1 (DATE) | F1 (LOC) | F1 (ORG) | F1 (PER) |
| -------------------------------------------------- | -------------------- | -------------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- |
| [xlm-roberta-base-finetuned-ner-luo](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-ner-luo) (This model) | [base](https://huggingface.co/xlm-roberta-base) | luo | 75.99 | 76.18 | 75.80 | 71.00 | 76.00 | 62.00 | 85.00 |
| [xlm-roberta-base-finetuned-luo-finetuned-ner-luo](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-luo-finetuned-ner-luo) | [luo](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-luo) | luo | 78.71 | 78.91 | 78.52 | 72.00 | 84.00 | 59.00 | 87.00 |
| [xlm-roberta-base-finetuned-swahili-finetuned-ner-luo](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-luo) | [swa](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-swahili) | luo | 78.13 | 77.75 | 78.52 | 65.00 | 82.00 | 61.00 | 89.00 |
## Usage
To use this model (or others), you can do the following, just changing the model name ([source](https://huggingface.co/dslim/bert-base-NER)):
```
from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
model_name = 'mbeukman/xlm-roberta-base-finetuned-ner-luo'
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
example = "Jii 2 moko jowito ngimagi ka machielo 1 to ohinyore marach mokalo e masira makoch mar apaya mane otimore e apaya mawuok Oyugis kochimo Chabera e sub county ma Rachuonyo East e County ma Homa Bay ewii odhiambo makawuononi"
ner_results = nlp(example)
print(ner_results)
```
|
mbeukman/xlm-roberta-base-finetuned-ner-luganda | mbeukman | 2021-11-25T09:04:33Z | 7 | 1 | transformers | [
"transformers",
"pytorch",
"xlm-roberta",
"token-classification",
"NER",
"lug",
"dataset:masakhaner",
"arxiv:2103.11811",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:05Z | ---
language:
- lug
tags:
- NER
datasets:
- masakhaner
metrics:
- f1
- precision
- recall
widget:
- text: "Empaka zaakubeera mu kibuga Liverpool e Bungereza , okutandika nga July 12 ."
---
# xlm-roberta-base-finetuned-ner-luganda
This is a token classification (specifically NER) model that fine-tuned [xlm-roberta-base](https://huggingface.co/xlm-roberta-base) on the [MasakhaNER](https://arxiv.org/abs/2103.11811) dataset, specifically the luganda part.
More information, and other similar models can be found in the [main Github repository](https://github.com/Michael-Beukman/NERTransfer).
## About
This model is transformer based and was fine-tuned on the MasakhaNER dataset. It is a named entity recognition dataset, containing mostly news articles in 10 different African languages.
The model was fine-tuned for 50 epochs, with a maximum sequence length of 200, 32 batch size, 5e-5 learning rate. This process was repeated 5 times (with different random seeds), and this uploaded model performed the best out of those 5 seeds (aggregate F1 on test set).
This model was fine-tuned by me, Michael Beukman while doing a project at the University of the Witwatersrand, Johannesburg. This is version 1, as of 20 November 2021.
This model is licensed under the [Apache License, Version 2.0](https://www.apache.org/licenses/LICENSE-2.0).
### Contact & More information
For more information about the models, including training scripts, detailed results and further resources, you can visit the the [main Github repository](https://github.com/Michael-Beukman/NERTransfer). You can contact me by filing an issue on this repository.
### Training Resources
In the interest of openness, and reporting resources used, we list here how long the training process took, as well as what the minimum resources would be to reproduce this. Fine-tuning each model on the NER dataset took between 10 and 30 minutes, and was performed on a NVIDIA RTX3090 GPU. To use a batch size of 32, at least 14GB of GPU memory was required, although it was just possible to fit these models in around 6.5GB's of VRAM when using a batch size of 1.
## Data
The train, evaluation and test datasets were taken directly from the MasakhaNER [Github](https://github.com/masakhane-io/masakhane-ner) repository, with minimal to no preprocessing, as the original dataset is already of high quality.
The motivation for the use of this data is that it is the "first large, publicly available, high quality dataset for named entity recognition (NER) in ten African languages" ([source](https://arxiv.org/pdf/2103.11811.pdf)). The high-quality data, as well as the groundwork laid by the paper introducing it are some more reasons why this dataset was used. For evaluation, the dedicated test split was used, which is from the same distribution as the training data, so this model may not generalise to other distributions, and further testing would need to be done to investigate this. The exact distribution of the data is covered in detail [here](https://arxiv.org/abs/2103.11811).
## Intended Use
This model are intended to be used for NLP research into e.g. interpretability or transfer learning. Using this model in production is not supported, as generalisability and downright performance is limited. In particular, this is not designed to be used in any important downstream task that could affect people, as harm could be caused by the limitations of the model, described next.
## Limitations
This model was only trained on one (relatively small) dataset, covering one task (NER) in one domain (news articles) and in a set span of time. The results may not generalise, and the model may perform badly, or in an unfair / biased way if used on other tasks. Although the purpose of this project was to investigate transfer learning, the performance on languages that the model was not trained for does suffer.
Because this model used xlm-roberta-base as its starting point (potentially with domain adaptive fine-tuning on specific languages), this model's limitations can also apply here. These can include being biased towards the hegemonic viewpoint of most of its training data, being ungrounded and having subpar results on other languages (possibly due to unbalanced training data).
As [Adelani et al. (2021)](https://arxiv.org/abs/2103.11811) showed, the models in general struggled with entities that were either longer than 3 words and entities that were not contained in the training data. This could bias the models towards not finding, e.g. names of people that have many words, possibly leading to a misrepresentation in the results. Similarly, names that are uncommon, and may not have been found in the training data (due to e.g. different languages) would also be predicted less often.
Additionally, this model has not been verified in practice, and other, more subtle problems may become prevalent if used without any verification that it does what it is supposed to.
### Privacy & Ethical Considerations
The data comes from only publicly available news sources, the only available data should cover public figures and those that agreed to be reported on. See the original MasakhaNER paper for more details.
No explicit ethical considerations or adjustments were made during fine-tuning of this model.
## Metrics
The language adaptive models achieve (mostly) superior performance over starting with xlm-roberta-base. Our main metric was the aggregate F1 score for all NER categories.
These metrics are on the test set for MasakhaNER, so the data distribution is similar to the training set, so these results do not directly indicate how well these models generalise.
We do find large variation in transfer results when starting from different seeds (5 different seeds were tested), indicating that the fine-tuning process for transfer might be unstable.
The metrics used were chosen to be consistent with previous work, and to facilitate research. Other metrics may be more appropriate for other purposes.
## Caveats and Recommendations
In general, this model performed worse on the 'date' category compared to others, so if dates are a critical factor, then that might need to be taken into account and addressed, by for example collecting and annotating more data.
## Model Structure
Here are some performance details on this specific model, compared to others we trained.
All of these metrics were calculated on the test set, and the seed was chosen that gave the best overall F1 score. The first three result columns are averaged over all categories, and the latter 4 provide performance broken down by category.
This model can predict the following label for a token ([source](https://huggingface.co/Davlan/xlm-roberta-large-masakhaner)):
Abbreviation|Description
-|-
O|Outside of a named entity
B-DATE |Beginning of a DATE entity right after another DATE entity
I-DATE |DATE entity
B-PER |Beginning of a person’s name right after another person’s name
I-PER |Person’s name
B-ORG |Beginning of an organisation right after another organisation
I-ORG |Organisation
B-LOC |Beginning of a location right after another location
I-LOC |Location
| Model Name | Staring point | Evaluation / Fine-tune Language | F1 | Precision | Recall | F1 (DATE) | F1 (LOC) | F1 (ORG) | F1 (PER) |
| -------------------------------------------------- | -------------------- | -------------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- |
| [xlm-roberta-base-finetuned-ner-luganda](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-ner-luganda) (This model) | [base](https://huggingface.co/xlm-roberta-base) | lug | 80.91 | 78.59 | 83.37 | 73.00 | 78.00 | 77.00 | 86.00 |
| [xlm-roberta-base-finetuned-luganda-finetuned-ner-luganda](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-luganda-finetuned-ner-luganda) | [lug](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-luganda) | lug | 85.37 | 82.75 | 88.17 | 78.00 | 82.00 | 80.00 | 92.00 |
| [xlm-roberta-base-finetuned-swahili-finetuned-ner-luganda](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-luganda) | [swa](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-swahili) | lug | 82.57 | 80.38 | 84.89 | 75.00 | 80.00 | 82.00 | 87.00 |
## Usage
To use this model (or others), you can do the following, just changing the model name ([source](https://huggingface.co/dslim/bert-base-NER)):
```
from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
model_name = 'mbeukman/xlm-roberta-base-finetuned-ner-luganda'
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
example = "Empaka zaakubeera mu kibuga Liverpool e Bungereza , okutandika nga July 12 ."
ner_results = nlp(example)
print(ner_results)
```
|
mbeukman/xlm-roberta-base-finetuned-ner-kinyarwanda | mbeukman | 2021-11-25T09:04:30Z | 8 | 0 | transformers | [
"transformers",
"pytorch",
"xlm-roberta",
"token-classification",
"NER",
"rw",
"dataset:masakhaner",
"arxiv:2103.11811",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:05Z | ---
language:
- rw
tags:
- NER
datasets:
- masakhaner
metrics:
- f1
- precision
- recall
widget:
- text: "Ambasaderi wa EU mu Rwanda , Nicola Bellomo yagize ati “ Inkunga yacu ni imwe mu nkunga yagutse yiswe # TeamEurope ."
---
# xlm-roberta-base-finetuned-ner-kinyarwanda
This is a token classification (specifically NER) model that fine-tuned [xlm-roberta-base](https://huggingface.co/xlm-roberta-base) on the [MasakhaNER](https://arxiv.org/abs/2103.11811) dataset, specifically the Kinyarwanda part.
More information, and other similar models can be found in the [main Github repository](https://github.com/Michael-Beukman/NERTransfer).
## About
This model is transformer based and was fine-tuned on the MasakhaNER dataset. It is a named entity recognition dataset, containing mostly news articles in 10 different African languages.
The model was fine-tuned for 50 epochs, with a maximum sequence length of 200, 32 batch size, 5e-5 learning rate. This process was repeated 5 times (with different random seeds), and this uploaded model performed the best out of those 5 seeds (aggregate F1 on test set).
This model was fine-tuned by me, Michael Beukman while doing a project at the University of the Witwatersrand, Johannesburg. This is version 1, as of 20 November 2021.
This model is licensed under the [Apache License, Version 2.0](https://www.apache.org/licenses/LICENSE-2.0).
### Contact & More information
For more information about the models, including training scripts, detailed results and further resources, you can visit the the [main Github repository](https://github.com/Michael-Beukman/NERTransfer). You can contact me by filing an issue on this repository.
### Training Resources
In the interest of openness, and reporting resources used, we list here how long the training process took, as well as what the minimum resources would be to reproduce this. Fine-tuning each model on the NER dataset took between 10 and 30 minutes, and was performed on a NVIDIA RTX3090 GPU. To use a batch size of 32, at least 14GB of GPU memory was required, although it was just possible to fit these models in around 6.5GB's of VRAM when using a batch size of 1.
## Data
The train, evaluation and test datasets were taken directly from the MasakhaNER [Github](https://github.com/masakhane-io/masakhane-ner) repository, with minimal to no preprocessing, as the original dataset is already of high quality.
The motivation for the use of this data is that it is the "first large, publicly available, high quality dataset for named entity recognition (NER) in ten African languages" ([source](https://arxiv.org/pdf/2103.11811.pdf)). The high-quality data, as well as the groundwork laid by the paper introducing it are some more reasons why this dataset was used. For evaluation, the dedicated test split was used, which is from the same distribution as the training data, so this model may not generalise to other distributions, and further testing would need to be done to investigate this. The exact distribution of the data is covered in detail [here](https://arxiv.org/abs/2103.11811).
## Intended Use
This model are intended to be used for NLP research into e.g. interpretability or transfer learning. Using this model in production is not supported, as generalisability and downright performance is limited. In particular, this is not designed to be used in any important downstream task that could affect people, as harm could be caused by the limitations of the model, described next.
## Limitations
This model was only trained on one (relatively small) dataset, covering one task (NER) in one domain (news articles) and in a set span of time. The results may not generalise, and the model may perform badly, or in an unfair / biased way if used on other tasks. Although the purpose of this project was to investigate transfer learning, the performance on languages that the model was not trained for does suffer.
Because this model used xlm-roberta-base as its starting point (potentially with domain adaptive fine-tuning on specific languages), this model's limitations can also apply here. These can include being biased towards the hegemonic viewpoint of most of its training data, being ungrounded and having subpar results on other languages (possibly due to unbalanced training data).
As [Adelani et al. (2021)](https://arxiv.org/abs/2103.11811) showed, the models in general struggled with entities that were either longer than 3 words and entities that were not contained in the training data. This could bias the models towards not finding, e.g. names of people that have many words, possibly leading to a misrepresentation in the results. Similarly, names that are uncommon, and may not have been found in the training data (due to e.g. different languages) would also be predicted less often.
Additionally, this model has not been verified in practice, and other, more subtle problems may become prevalent if used without any verification that it does what it is supposed to.
### Privacy & Ethical Considerations
The data comes from only publicly available news sources, the only available data should cover public figures and those that agreed to be reported on. See the original MasakhaNER paper for more details.
No explicit ethical considerations or adjustments were made during fine-tuning of this model.
## Metrics
The language adaptive models achieve (mostly) superior performance over starting with xlm-roberta-base. Our main metric was the aggregate F1 score for all NER categories.
These metrics are on the test set for MasakhaNER, so the data distribution is similar to the training set, so these results do not directly indicate how well these models generalise.
We do find large variation in transfer results when starting from different seeds (5 different seeds were tested), indicating that the fine-tuning process for transfer might be unstable.
The metrics used were chosen to be consistent with previous work, and to facilitate research. Other metrics may be more appropriate for other purposes.
## Caveats and Recommendations
In general, this model performed worse on the 'date' category compared to others, so if dates are a critical factor, then that might need to be taken into account and addressed, by for example collecting and annotating more data.
## Model Structure
Here are some performance details on this specific model, compared to others we trained.
All of these metrics were calculated on the test set, and the seed was chosen that gave the best overall F1 score. The first three result columns are averaged over all categories, and the latter 4 provide performance broken down by category.
This model can predict the following label for a token ([source](https://huggingface.co/Davlan/xlm-roberta-large-masakhaner)):
Abbreviation|Description
-|-
O|Outside of a named entity
B-DATE |Beginning of a DATE entity right after another DATE entity
I-DATE |DATE entity
B-PER |Beginning of a person’s name right after another person’s name
I-PER |Person’s name
B-ORG |Beginning of an organisation right after another organisation
I-ORG |Organisation
B-LOC |Beginning of a location right after another location
I-LOC |Location
| Model Name | Staring point | Evaluation / Fine-tune Language | F1 | Precision | Recall | F1 (DATE) | F1 (LOC) | F1 (ORG) | F1 (PER) |
| -------------------------------------------------- | -------------------- | -------------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- |
| [xlm-roberta-base-finetuned-ner-kinyarwanda](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-ner-kinyarwanda) (This model) | [base](https://huggingface.co/xlm-roberta-base) | kin | 74.59 | 72.17 | 77.17 | 70.00 | 75.00 | 70.00 | 82.00 |
| [xlm-roberta-base-finetuned-kinyarwanda-finetuned-ner-kinyarwanda](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-kinyarwanda-finetuned-ner-kinyarwanda) | [kin](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-kinyarwanda) | kin | 79.55 | 75.56 | 83.99 | 69.00 | 79.00 | 77.00 | 90.00 |
| [xlm-roberta-base-finetuned-swahili-finetuned-ner-kinyarwanda](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-kinyarwanda) | [swa](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-swahili) | kin | 76.31 | 72.64 | 80.37 | 70.00 | 76.00 | 75.00 | 84.00 |
## Usage
To use this model (or others), you can do the following, just changing the model name ([source](https://huggingface.co/dslim/bert-base-NER)):
```
from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
model_name = 'mbeukman/xlm-roberta-base-finetuned-ner-kinyarwanda'
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
example = "Ambasaderi wa EU mu Rwanda , Nicola Bellomo yagize ati “ Inkunga yacu ni imwe mu nkunga yagutse yiswe # TeamEurope ."
ner_results = nlp(example)
print(ner_results)
```
|
mbeukman/xlm-roberta-base-finetuned-ner-igbo | mbeukman | 2021-11-25T09:04:28Z | 7 | 0 | transformers | [
"transformers",
"pytorch",
"xlm-roberta",
"token-classification",
"NER",
"ig",
"dataset:masakhaner",
"arxiv:2103.11811",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:05Z | ---
language:
- ig
tags:
- NER
datasets:
- masakhaner
metrics:
- f1
- precision
- recall
widget:
- text: "Ike ịda jụụ otụ nkeji banyere oke ogbugbu na - eme n'ala Naijiria agwụla Ekweremmadụ"
---
# xlm-roberta-base-finetuned-ner-igbo
This is a token classification (specifically NER) model that fine-tuned [xlm-roberta-base](https://huggingface.co/xlm-roberta-base) on the [MasakhaNER](https://arxiv.org/abs/2103.11811) dataset, specifically the Igbo part.
More information, and other similar models can be found in the [main Github repository](https://github.com/Michael-Beukman/NERTransfer).
## About
This model is transformer based and was fine-tuned on the MasakhaNER dataset. It is a named entity recognition dataset, containing mostly news articles in 10 different African languages.
The model was fine-tuned for 50 epochs, with a maximum sequence length of 200, 32 batch size, 5e-5 learning rate. This process was repeated 5 times (with different random seeds), and this uploaded model performed the best out of those 5 seeds (aggregate F1 on test set).
This model was fine-tuned by me, Michael Beukman while doing a project at the University of the Witwatersrand, Johannesburg. This is version 1, as of 20 November 2021.
This model is licensed under the [Apache License, Version 2.0](https://www.apache.org/licenses/LICENSE-2.0).
### Contact & More information
For more information about the models, including training scripts, detailed results and further resources, you can visit the the [main Github repository](https://github.com/Michael-Beukman/NERTransfer). You can contact me by filing an issue on this repository.
### Training Resources
In the interest of openness, and reporting resources used, we list here how long the training process took, as well as what the minimum resources would be to reproduce this. Fine-tuning each model on the NER dataset took between 10 and 30 minutes, and was performed on a NVIDIA RTX3090 GPU. To use a batch size of 32, at least 14GB of GPU memory was required, although it was just possible to fit these models in around 6.5GB's of VRAM when using a batch size of 1.
## Data
The train, evaluation and test datasets were taken directly from the MasakhaNER [Github](https://github.com/masakhane-io/masakhane-ner) repository, with minimal to no preprocessing, as the original dataset is already of high quality.
The motivation for the use of this data is that it is the "first large, publicly available, high quality dataset for named entity recognition (NER) in ten African languages" ([source](https://arxiv.org/pdf/2103.11811.pdf)). The high-quality data, as well as the groundwork laid by the paper introducing it are some more reasons why this dataset was used. For evaluation, the dedicated test split was used, which is from the same distribution as the training data, so this model may not generalise to other distributions, and further testing would need to be done to investigate this. The exact distribution of the data is covered in detail [here](https://arxiv.org/abs/2103.11811).
## Intended Use
This model are intended to be used for NLP research into e.g. interpretability or transfer learning. Using this model in production is not supported, as generalisability and downright performance is limited. In particular, this is not designed to be used in any important downstream task that could affect people, as harm could be caused by the limitations of the model, described next.
## Limitations
This model was only trained on one (relatively small) dataset, covering one task (NER) in one domain (news articles) and in a set span of time. The results may not generalise, and the model may perform badly, or in an unfair / biased way if used on other tasks. Although the purpose of this project was to investigate transfer learning, the performance on languages that the model was not trained for does suffer.
Because this model used xlm-roberta-base as its starting point (potentially with domain adaptive fine-tuning on specific languages), this model's limitations can also apply here. These can include being biased towards the hegemonic viewpoint of most of its training data, being ungrounded and having subpar results on other languages (possibly due to unbalanced training data).
As [Adelani et al. (2021)](https://arxiv.org/abs/2103.11811) showed, the models in general struggled with entities that were either longer than 3 words and entities that were not contained in the training data. This could bias the models towards not finding, e.g. names of people that have many words, possibly leading to a misrepresentation in the results. Similarly, names that are uncommon, and may not have been found in the training data (due to e.g. different languages) would also be predicted less often.
Additionally, this model has not been verified in practice, and other, more subtle problems may become prevalent if used without any verification that it does what it is supposed to.
### Privacy & Ethical Considerations
The data comes from only publicly available news sources, the only available data should cover public figures and those that agreed to be reported on. See the original MasakhaNER paper for more details.
No explicit ethical considerations or adjustments were made during fine-tuning of this model.
## Metrics
The language adaptive models achieve (mostly) superior performance over starting with xlm-roberta-base. Our main metric was the aggregate F1 score for all NER categories.
These metrics are on the test set for MasakhaNER, so the data distribution is similar to the training set, so these results do not directly indicate how well these models generalise.
We do find large variation in transfer results when starting from different seeds (5 different seeds were tested), indicating that the fine-tuning process for transfer might be unstable.
The metrics used were chosen to be consistent with previous work, and to facilitate research. Other metrics may be more appropriate for other purposes.
## Caveats and Recommendations
In general, this model performed worse on the 'date' category compared to others, so if dates are a critical factor, then that might need to be taken into account and addressed, by for example collecting and annotating more data.
## Model Structure
Here are some performance details on this specific model, compared to others we trained.
All of these metrics were calculated on the test set, and the seed was chosen that gave the best overall F1 score. The first three result columns are averaged over all categories, and the latter 4 provide performance broken down by category.
This model can predict the following label for a token ([source](https://huggingface.co/Davlan/xlm-roberta-large-masakhaner)):
Abbreviation|Description
-|-
O|Outside of a named entity
B-DATE |Beginning of a DATE entity right after another DATE entity
I-DATE |DATE entity
B-PER |Beginning of a person’s name right after another person’s name
I-PER |Person’s name
B-ORG |Beginning of an organisation right after another organisation
I-ORG |Organisation
B-LOC |Beginning of a location right after another location
I-LOC |Location
| Model Name | Staring point | Evaluation / Fine-tune Language | F1 | Precision | Recall | F1 (DATE) | F1 (LOC) | F1 (ORG) | F1 (PER) |
| -------------------------------------------------- | -------------------- | -------------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- |
| [xlm-roberta-base-finetuned-ner-igbo](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-ner-igbo) (This model) | [base](https://huggingface.co/xlm-roberta-base) | ibo | 86.06 | 85.20 | 86.94 | 76.00 | 86.00 | 90.00 | 87.00 |
| [xlm-roberta-base-finetuned-igbo-finetuned-ner-igbo](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-igbo-finetuned-ner-igbo) | [ibo](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-igbo) | ibo | 88.39 | 87.08 | 89.74 | 74.00 | 91.00 | 90.00 | 91.00 |
| [xlm-roberta-base-finetuned-swahili-finetuned-ner-igbo](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-igbo) | [swa](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-swahili) | ibo | 84.93 | 83.63 | 86.26 | 70.00 | 88.00 | 89.00 | 84.00 |
## Usage
To use this model (or others), you can do the following, just changing the model name ([source](https://huggingface.co/dslim/bert-base-NER)):
```
from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
model_name = 'mbeukman/xlm-roberta-base-finetuned-ner-igbo'
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
example = "Ike ịda jụụ otụ nkeji banyere oke ogbugbu na - eme n'ala Naijiria agwụla Ekweremmadụ"
ner_results = nlp(example)
print(ner_results)
```
|
mbeukman/xlm-roberta-base-finetuned-ner-hausa | mbeukman | 2021-11-25T09:04:25Z | 7 | 0 | transformers | [
"transformers",
"pytorch",
"xlm-roberta",
"token-classification",
"NER",
"ha",
"dataset:masakhaner",
"arxiv:2103.11811",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:05Z | ---
language:
- ha
tags:
- NER
datasets:
- masakhaner
metrics:
- f1
- precision
- recall
widget:
- text: "A saurari cikakken rahoton wakilin Muryar Amurka Ibrahim Abdul'aziz"
---
# xlm-roberta-base-finetuned-ner-hausa
This is a token classification (specifically NER) model that fine-tuned [xlm-roberta-base](https://huggingface.co/xlm-roberta-base) on the [MasakhaNER](https://arxiv.org/abs/2103.11811) dataset, specifically the Hausa part.
More information, and other similar models can be found in the [main Github repository](https://github.com/Michael-Beukman/NERTransfer).
## About
This model is transformer based and was fine-tuned on the MasakhaNER dataset. It is a named entity recognition dataset, containing mostly news articles in 10 different African languages.
The model was fine-tuned for 50 epochs, with a maximum sequence length of 200, 32 batch size, 5e-5 learning rate. This process was repeated 5 times (with different random seeds), and this uploaded model performed the best out of those 5 seeds (aggregate F1 on test set).
This model was fine-tuned by me, Michael Beukman while doing a project at the University of the Witwatersrand, Johannesburg. This is version 1, as of 20 November 2021.
This model is licensed under the [Apache License, Version 2.0](https://www.apache.org/licenses/LICENSE-2.0).
### Contact & More information
For more information about the models, including training scripts, detailed results and further resources, you can visit the the [main Github repository](https://github.com/Michael-Beukman/NERTransfer). You can contact me by filing an issue on this repository.
### Training Resources
In the interest of openness, and reporting resources used, we list here how long the training process took, as well as what the minimum resources would be to reproduce this. Fine-tuning each model on the NER dataset took between 10 and 30 minutes, and was performed on a NVIDIA RTX3090 GPU. To use a batch size of 32, at least 14GB of GPU memory was required, although it was just possible to fit these models in around 6.5GB's of VRAM when using a batch size of 1.
## Data
The train, evaluation and test datasets were taken directly from the MasakhaNER [Github](https://github.com/masakhane-io/masakhane-ner) repository, with minimal to no preprocessing, as the original dataset is already of high quality.
The motivation for the use of this data is that it is the "first large, publicly available, high quality dataset for named entity recognition (NER) in ten African languages" ([source](https://arxiv.org/pdf/2103.11811.pdf)). The high-quality data, as well as the groundwork laid by the paper introducing it are some more reasons why this dataset was used. For evaluation, the dedicated test split was used, which is from the same distribution as the training data, so this model may not generalise to other distributions, and further testing would need to be done to investigate this. The exact distribution of the data is covered in detail [here](https://arxiv.org/abs/2103.11811).
## Intended Use
This model are intended to be used for NLP research into e.g. interpretability or transfer learning. Using this model in production is not supported, as generalisability and downright performance is limited. In particular, this is not designed to be used in any important downstream task that could affect people, as harm could be caused by the limitations of the model, described next.
## Limitations
This model was only trained on one (relatively small) dataset, covering one task (NER) in one domain (news articles) and in a set span of time. The results may not generalise, and the model may perform badly, or in an unfair / biased way if used on other tasks. Although the purpose of this project was to investigate transfer learning, the performance on languages that the model was not trained for does suffer.
Because this model used xlm-roberta-base as its starting point (potentially with domain adaptive fine-tuning on specific languages), this model's limitations can also apply here. These can include being biased towards the hegemonic viewpoint of most of its training data, being ungrounded and having subpar results on other languages (possibly due to unbalanced training data).
As [Adelani et al. (2021)](https://arxiv.org/abs/2103.11811) showed, the models in general struggled with entities that were either longer than 3 words and entities that were not contained in the training data. This could bias the models towards not finding, e.g. names of people that have many words, possibly leading to a misrepresentation in the results. Similarly, names that are uncommon, and may not have been found in the training data (due to e.g. different languages) would also be predicted less often.
Additionally, this model has not been verified in practice, and other, more subtle problems may become prevalent if used without any verification that it does what it is supposed to.
### Privacy & Ethical Considerations
The data comes from only publicly available news sources, the only available data should cover public figures and those that agreed to be reported on. See the original MasakhaNER paper for more details.
No explicit ethical considerations or adjustments were made during fine-tuning of this model.
## Metrics
The language adaptive models achieve (mostly) superior performance over starting with xlm-roberta-base. Our main metric was the aggregate F1 score for all NER categories.
These metrics are on the test set for MasakhaNER, so the data distribution is similar to the training set, so these results do not directly indicate how well these models generalise.
We do find large variation in transfer results when starting from different seeds (5 different seeds were tested), indicating that the fine-tuning process for transfer might be unstable.
The metrics used were chosen to be consistent with previous work, and to facilitate research. Other metrics may be more appropriate for other purposes.
## Caveats and Recommendations
In general, this model performed worse on the 'date' category compared to others, so if dates are a critical factor, then that might need to be taken into account and addressed, by for example collecting and annotating more data.
## Model Structure
Here are some performance details on this specific model, compared to others we trained.
All of these metrics were calculated on the test set, and the seed was chosen that gave the best overall F1 score. The first three result columns are averaged over all categories, and the latter 4 provide performance broken down by category.
This model can predict the following label for a token ([source](https://huggingface.co/Davlan/xlm-roberta-large-masakhaner)):
Abbreviation|Description
-|-
O|Outside of a named entity
B-DATE |Beginning of a DATE entity right after another DATE entity
I-DATE |DATE entity
B-PER |Beginning of a person’s name right after another person’s name
I-PER |Person’s name
B-ORG |Beginning of an organisation right after another organisation
I-ORG |Organisation
B-LOC |Beginning of a location right after another location
I-LOC |Location
| Model Name | Staring point | Evaluation / Fine-tune Language | F1 | Precision | Recall | F1 (DATE) | F1 (LOC) | F1 (ORG) | F1 (PER) |
| -------------------------------------------------- | -------------------- | -------------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- |
| [xlm-roberta-base-finetuned-ner-hausa](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-ner-hausa) (This model) | [base](https://huggingface.co/xlm-roberta-base) | hau | 89.94 | 87.74 | 92.25 | 84.00 | 94.00 | 74.00 | 93.00 |
| [xlm-roberta-base-finetuned-hausa-finetuned-ner-hausa](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-hausa-finetuned-ner-hausa) | [hau](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-hausa) | hau | 92.27 | 90.46 | 94.16 | 85.00 | 95.00 | 80.00 | 97.00 |
| [xlm-roberta-base-finetuned-swahili-finetuned-ner-hausa](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-hausa) | [swa](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-swahili) | hau | 89.14 | 87.18 | 91.20 | 82.00 | 93.00 | 76.00 | 93.00 |
## Usage
To use this model (or others), you can do the following, just changing the model name ([source](https://huggingface.co/dslim/bert-base-NER)):
```
from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
model_name = 'mbeukman/xlm-roberta-base-finetuned-ner-hausa'
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
example = "A saurari cikakken rahoton wakilin Muryar Amurka Ibrahim Abdul'aziz"
ner_results = nlp(example)
print(ner_results)
```
|
mbeukman/xlm-roberta-base-finetuned-naija-finetuned-ner-naija | mbeukman | 2021-11-25T09:04:20Z | 4 | 0 | transformers | [
"transformers",
"pytorch",
"xlm-roberta",
"token-classification",
"NER",
"pcm",
"dataset:masakhaner",
"arxiv:2103.11811",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:05Z | ---
language:
- pcm
tags:
- NER
datasets:
- masakhaner
metrics:
- f1
- precision
- recall
widget:
- text: "Mixed Martial Arts joinbodi , Ultimate Fighting Championship , UFC don decide say dem go enta back di octagon on Saturday , 9 May , for Jacksonville , Florida ."
---
# xlm-roberta-base-finetuned-naija-finetuned-ner-naija
This is a token classification (specifically NER) model that fine-tuned [xlm-roberta-base-finetuned-naija](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-naija) on the [MasakhaNER](https://arxiv.org/abs/2103.11811) dataset, specifically the Nigerian Pidgin part.
More information, and other similar models can be found in the [main Github repository](https://github.com/Michael-Beukman/NERTransfer).
## About
This model is transformer based and was fine-tuned on the MasakhaNER dataset. It is a named entity recognition dataset, containing mostly news articles in 10 different African languages.
The model was fine-tuned for 50 epochs, with a maximum sequence length of 200, 32 batch size, 5e-5 learning rate. This process was repeated 5 times (with different random seeds), and this uploaded model performed the best out of those 5 seeds (aggregate F1 on test set).
This model was fine-tuned by me, Michael Beukman while doing a project at the University of the Witwatersrand, Johannesburg. This is version 1, as of 20 November 2021.
This model is licensed under the [Apache License, Version 2.0](https://www.apache.org/licenses/LICENSE-2.0).
### Contact & More information
For more information about the models, including training scripts, detailed results and further resources, you can visit the the [main Github repository](https://github.com/Michael-Beukman/NERTransfer). You can contact me by filing an issue on this repository.
### Training Resources
In the interest of openness, and reporting resources used, we list here how long the training process took, as well as what the minimum resources would be to reproduce this. Fine-tuning each model on the NER dataset took between 10 and 30 minutes, and was performed on a NVIDIA RTX3090 GPU. To use a batch size of 32, at least 14GB of GPU memory was required, although it was just possible to fit these models in around 6.5GB's of VRAM when using a batch size of 1.
## Data
The train, evaluation and test datasets were taken directly from the MasakhaNER [Github](https://github.com/masakhane-io/masakhane-ner) repository, with minimal to no preprocessing, as the original dataset is already of high quality.
The motivation for the use of this data is that it is the "first large, publicly available, high quality dataset for named entity recognition (NER) in ten African languages" ([source](https://arxiv.org/pdf/2103.11811.pdf)). The high-quality data, as well as the groundwork laid by the paper introducing it are some more reasons why this dataset was used. For evaluation, the dedicated test split was used, which is from the same distribution as the training data, so this model may not generalise to other distributions, and further testing would need to be done to investigate this. The exact distribution of the data is covered in detail [here](https://arxiv.org/abs/2103.11811).
## Intended Use
This model are intended to be used for NLP research into e.g. interpretability or transfer learning. Using this model in production is not supported, as generalisability and downright performance is limited. In particular, this is not designed to be used in any important downstream task that could affect people, as harm could be caused by the limitations of the model, described next.
## Limitations
This model was only trained on one (relatively small) dataset, covering one task (NER) in one domain (news articles) and in a set span of time. The results may not generalise, and the model may perform badly, or in an unfair / biased way if used on other tasks. Although the purpose of this project was to investigate transfer learning, the performance on languages that the model was not trained for does suffer.
Because this model used xlm-roberta-base as its starting point (potentially with domain adaptive fine-tuning on specific languages), this model's limitations can also apply here. These can include being biased towards the hegemonic viewpoint of most of its training data, being ungrounded and having subpar results on other languages (possibly due to unbalanced training data).
As [Adelani et al. (2021)](https://arxiv.org/abs/2103.11811) showed, the models in general struggled with entities that were either longer than 3 words and entities that were not contained in the training data. This could bias the models towards not finding, e.g. names of people that have many words, possibly leading to a misrepresentation in the results. Similarly, names that are uncommon, and may not have been found in the training data (due to e.g. different languages) would also be predicted less often.
Additionally, this model has not been verified in practice, and other, more subtle problems may become prevalent if used without any verification that it does what it is supposed to.
### Privacy & Ethical Considerations
The data comes from only publicly available news sources, the only available data should cover public figures and those that agreed to be reported on. See the original MasakhaNER paper for more details.
No explicit ethical considerations or adjustments were made during fine-tuning of this model.
## Metrics
The language adaptive models achieve (mostly) superior performance over starting with xlm-roberta-base. Our main metric was the aggregate F1 score for all NER categories.
These metrics are on the test set for MasakhaNER, so the data distribution is similar to the training set, so these results do not directly indicate how well these models generalise.
We do find large variation in transfer results when starting from different seeds (5 different seeds were tested), indicating that the fine-tuning process for transfer might be unstable.
The metrics used were chosen to be consistent with previous work, and to facilitate research. Other metrics may be more appropriate for other purposes.
## Caveats and Recommendations
In general, this model performed worse on the 'date' category compared to others, so if dates are a critical factor, then that might need to be taken into account and addressed, by for example collecting and annotating more data.
## Model Structure
Here are some performance details on this specific model, compared to others we trained.
All of these metrics were calculated on the test set, and the seed was chosen that gave the best overall F1 score. The first three result columns are averaged over all categories, and the latter 4 provide performance broken down by category.
This model can predict the following label for a token ([source](https://huggingface.co/Davlan/xlm-roberta-large-masakhaner)):
Abbreviation|Description
-|-
O|Outside of a named entity
B-DATE |Beginning of a DATE entity right after another DATE entity
I-DATE |DATE entity
B-PER |Beginning of a person’s name right after another person’s name
I-PER |Person’s name
B-ORG |Beginning of an organisation right after another organisation
I-ORG |Organisation
B-LOC |Beginning of a location right after another location
I-LOC |Location
| Model Name | Staring point | Evaluation / Fine-tune Language | F1 | Precision | Recall | F1 (DATE) | F1 (LOC) | F1 (ORG) | F1 (PER) |
| -------------------------------------------------- | -------------------- | -------------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- |
| [xlm-roberta-base-finetuned-naija-finetuned-ner-naija](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-naija-finetuned-ner-naija) (This model) | [pcm](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-naija) | pcm | 88.06 | 87.04 | 89.12 | 90.00 | 88.00 | 81.00 | 92.00 |
| [xlm-roberta-base-finetuned-swahili-finetuned-ner-naija](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-naija) | [swa](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-swahili) | pcm | 89.12 | 87.84 | 90.42 | 90.00 | 89.00 | 82.00 | 94.00 |
| [xlm-roberta-base-finetuned-ner-naija](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-ner-naija) | [base](https://huggingface.co/xlm-roberta-base) | pcm | 88.89 | 88.13 | 89.66 | 92.00 | 87.00 | 82.00 | 94.00 |
## Usage
To use this model (or others), you can do the following, just changing the model name ([source](https://huggingface.co/dslim/bert-base-NER)):
```
from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
model_name = 'mbeukman/xlm-roberta-base-finetuned-naija-finetuned-ner-naija'
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
example = "Mixed Martial Arts joinbodi , Ultimate Fighting Championship , UFC don decide say dem go enta back di octagon on Saturday , 9 May , for Jacksonville , Florida ."
ner_results = nlp(example)
print(ner_results)
```
|
mbeukman/xlm-roberta-base-finetuned-luo-finetuned-ner-luo | mbeukman | 2021-11-25T09:04:15Z | 18 | 0 | transformers | [
"transformers",
"pytorch",
"xlm-roberta",
"token-classification",
"NER",
"luo",
"dataset:masakhaner",
"arxiv:2103.11811",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:05Z | ---
language:
- luo
tags:
- NER
datasets:
- masakhaner
metrics:
- f1
- precision
- recall
widget:
- text: "Jii 2 moko jowito ngimagi ka machielo 1 to ohinyore marach mokalo e masira makoch mar apaya mane otimore e apaya mawuok Oyugis kochimo Chabera e sub county ma Rachuonyo East e County ma Homa Bay ewii odhiambo makawuononi"
---
# xlm-roberta-base-finetuned-luo-finetuned-ner-luo
This is a token classification (specifically NER) model that fine-tuned [xlm-roberta-base-finetuned-luo](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-luo) on the [MasakhaNER](https://arxiv.org/abs/2103.11811) dataset, specifically the Luo part.
More information, and other similar models can be found in the [main Github repository](https://github.com/Michael-Beukman/NERTransfer).
## About
This model is transformer based and was fine-tuned on the MasakhaNER dataset. It is a named entity recognition dataset, containing mostly news articles in 10 different African languages.
The model was fine-tuned for 50 epochs, with a maximum sequence length of 200, 32 batch size, 5e-5 learning rate. This process was repeated 5 times (with different random seeds), and this uploaded model performed the best out of those 5 seeds (aggregate F1 on test set).
This model was fine-tuned by me, Michael Beukman while doing a project at the University of the Witwatersrand, Johannesburg. This is version 1, as of 20 November 2021.
This model is licensed under the [Apache License, Version 2.0](https://www.apache.org/licenses/LICENSE-2.0).
### Contact & More information
For more information about the models, including training scripts, detailed results and further resources, you can visit the the [main Github repository](https://github.com/Michael-Beukman/NERTransfer). You can contact me by filing an issue on this repository.
### Training Resources
In the interest of openness, and reporting resources used, we list here how long the training process took, as well as what the minimum resources would be to reproduce this. Fine-tuning each model on the NER dataset took between 10 and 30 minutes, and was performed on a NVIDIA RTX3090 GPU. To use a batch size of 32, at least 14GB of GPU memory was required, although it was just possible to fit these models in around 6.5GB's of VRAM when using a batch size of 1.
## Data
The train, evaluation and test datasets were taken directly from the MasakhaNER [Github](https://github.com/masakhane-io/masakhane-ner) repository, with minimal to no preprocessing, as the original dataset is already of high quality.
The motivation for the use of this data is that it is the "first large, publicly available, high quality dataset for named entity recognition (NER) in ten African languages" ([source](https://arxiv.org/pdf/2103.11811.pdf)). The high-quality data, as well as the groundwork laid by the paper introducing it are some more reasons why this dataset was used. For evaluation, the dedicated test split was used, which is from the same distribution as the training data, so this model may not generalise to other distributions, and further testing would need to be done to investigate this. The exact distribution of the data is covered in detail [here](https://arxiv.org/abs/2103.11811).
## Intended Use
This model are intended to be used for NLP research into e.g. interpretability or transfer learning. Using this model in production is not supported, as generalisability and downright performance is limited. In particular, this is not designed to be used in any important downstream task that could affect people, as harm could be caused by the limitations of the model, described next.
## Limitations
This model was only trained on one (relatively small) dataset, covering one task (NER) in one domain (news articles) and in a set span of time. The results may not generalise, and the model may perform badly, or in an unfair / biased way if used on other tasks. Although the purpose of this project was to investigate transfer learning, the performance on languages that the model was not trained for does suffer.
Because this model used xlm-roberta-base as its starting point (potentially with domain adaptive fine-tuning on specific languages), this model's limitations can also apply here. These can include being biased towards the hegemonic viewpoint of most of its training data, being ungrounded and having subpar results on other languages (possibly due to unbalanced training data).
As [Adelani et al. (2021)](https://arxiv.org/abs/2103.11811) showed, the models in general struggled with entities that were either longer than 3 words and entities that were not contained in the training data. This could bias the models towards not finding, e.g. names of people that have many words, possibly leading to a misrepresentation in the results. Similarly, names that are uncommon, and may not have been found in the training data (due to e.g. different languages) would also be predicted less often.
Additionally, this model has not been verified in practice, and other, more subtle problems may become prevalent if used without any verification that it does what it is supposed to.
### Privacy & Ethical Considerations
The data comes from only publicly available news sources, the only available data should cover public figures and those that agreed to be reported on. See the original MasakhaNER paper for more details.
No explicit ethical considerations or adjustments were made during fine-tuning of this model.
## Metrics
The language adaptive models achieve (mostly) superior performance over starting with xlm-roberta-base. Our main metric was the aggregate F1 score for all NER categories.
These metrics are on the test set for MasakhaNER, so the data distribution is similar to the training set, so these results do not directly indicate how well these models generalise.
We do find large variation in transfer results when starting from different seeds (5 different seeds were tested), indicating that the fine-tuning process for transfer might be unstable.
The metrics used were chosen to be consistent with previous work, and to facilitate research. Other metrics may be more appropriate for other purposes.
## Caveats and Recommendations
In general, this model performed worse on the 'date' category compared to others, so if dates are a critical factor, then that might need to be taken into account and addressed, by for example collecting and annotating more data.
## Model Structure
Here are some performance details on this specific model, compared to others we trained.
All of these metrics were calculated on the test set, and the seed was chosen that gave the best overall F1 score. The first three result columns are averaged over all categories, and the latter 4 provide performance broken down by category.
This model can predict the following label for a token ([source](https://huggingface.co/Davlan/xlm-roberta-large-masakhaner)):
Abbreviation|Description
-|-
O|Outside of a named entity
B-DATE |Beginning of a DATE entity right after another DATE entity
I-DATE |DATE entity
B-PER |Beginning of a person’s name right after another person’s name
I-PER |Person’s name
B-ORG |Beginning of an organisation right after another organisation
I-ORG |Organisation
B-LOC |Beginning of a location right after another location
I-LOC |Location
| Model Name | Staring point | Evaluation / Fine-tune Language | F1 | Precision | Recall | F1 (DATE) | F1 (LOC) | F1 (ORG) | F1 (PER) |
| -------------------------------------------------- | -------------------- | -------------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- |
| [xlm-roberta-base-finetuned-luo-finetuned-ner-luo](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-luo-finetuned-ner-luo) (This model) | [luo](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-luo) | luo | 78.71 | 78.91 | 78.52 | 72.00 | 84.00 | 59.00 | 87.00 |
| [xlm-roberta-base-finetuned-swahili-finetuned-ner-luo](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-luo) | [swa](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-swahili) | luo | 78.13 | 77.75 | 78.52 | 65.00 | 82.00 | 61.00 | 89.00 |
| [xlm-roberta-base-finetuned-ner-luo](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-ner-luo) | [base](https://huggingface.co/xlm-roberta-base) | luo | 75.99 | 76.18 | 75.80 | 71.00 | 76.00 | 62.00 | 85.00 |
## Usage
To use this model (or others), you can do the following, just changing the model name ([source](https://huggingface.co/dslim/bert-base-NER)):
```
from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
model_name = 'mbeukman/xlm-roberta-base-finetuned-luo-finetuned-ner-luo'
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
example = "Jii 2 moko jowito ngimagi ka machielo 1 to ohinyore marach mokalo e masira makoch mar apaya mane otimore e apaya mawuok Oyugis kochimo Chabera e sub county ma Rachuonyo East e County ma Homa Bay ewii odhiambo makawuononi"
ner_results = nlp(example)
print(ner_results)
```
|
mbeukman/xlm-roberta-base-finetuned-igbo-finetuned-ner-swahili | mbeukman | 2021-11-25T09:04:02Z | 6 | 1 | transformers | [
"transformers",
"pytorch",
"xlm-roberta",
"token-classification",
"NER",
"sw",
"dataset:masakhaner",
"arxiv:2103.11811",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:05Z | ---
language:
- sw
tags:
- NER
datasets:
- masakhaner
metrics:
- f1
- precision
- recall
widget:
- text: "Wizara ya afya ya Tanzania imeripoti Jumatatu kuwa , watu takriban 14 zaidi wamepata maambukizi ya Covid - 19 ."
---
# xlm-roberta-base-finetuned-igbo-finetuned-ner-swahili
This is a token classification (specifically NER) model that fine-tuned [xlm-roberta-base-finetuned-igbo](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-igbo) on the [MasakhaNER](https://arxiv.org/abs/2103.11811) dataset, specifically the Swahili part.
More information, and other similar models can be found in the [main Github repository](https://github.com/Michael-Beukman/NERTransfer).
## About
This model is transformer based and was fine-tuned on the MasakhaNER dataset. It is a named entity recognition dataset, containing mostly news articles in 10 different African languages.
The model was fine-tuned for 50 epochs, with a maximum sequence length of 200, 32 batch size, 5e-5 learning rate. This process was repeated 5 times (with different random seeds), and this uploaded model performed the best out of those 5 seeds (aggregate F1 on test set).
This model was fine-tuned by me, Michael Beukman while doing a project at the University of the Witwatersrand, Johannesburg. This is version 1, as of 20 November 2021.
This model is licensed under the [Apache License, Version 2.0](https://www.apache.org/licenses/LICENSE-2.0).
### Contact & More information
For more information about the models, including training scripts, detailed results and further resources, you can visit the the [main Github repository](https://github.com/Michael-Beukman/NERTransfer). You can contact me by filing an issue on this repository.
### Training Resources
In the interest of openness, and reporting resources used, we list here how long the training process took, as well as what the minimum resources would be to reproduce this. Fine-tuning each model on the NER dataset took between 10 and 30 minutes, and was performed on a NVIDIA RTX3090 GPU. To use a batch size of 32, at least 14GB of GPU memory was required, although it was just possible to fit these models in around 6.5GB's of VRAM when using a batch size of 1.
## Data
The train, evaluation and test datasets were taken directly from the MasakhaNER [Github](https://github.com/masakhane-io/masakhane-ner) repository, with minimal to no preprocessing, as the original dataset is already of high quality.
The motivation for the use of this data is that it is the "first large, publicly available, high quality dataset for named entity recognition (NER) in ten African languages" ([source](https://arxiv.org/pdf/2103.11811.pdf)). The high-quality data, as well as the groundwork laid by the paper introducing it are some more reasons why this dataset was used. For evaluation, the dedicated test split was used, which is from the same distribution as the training data, so this model may not generalise to other distributions, and further testing would need to be done to investigate this. The exact distribution of the data is covered in detail [here](https://arxiv.org/abs/2103.11811).
## Intended Use
This model are intended to be used for NLP research into e.g. interpretability or transfer learning. Using this model in production is not supported, as generalisability and downright performance is limited. In particular, this is not designed to be used in any important downstream task that could affect people, as harm could be caused by the limitations of the model, described next.
## Limitations
This model was only trained on one (relatively small) dataset, covering one task (NER) in one domain (news articles) and in a set span of time. The results may not generalise, and the model may perform badly, or in an unfair / biased way if used on other tasks. Although the purpose of this project was to investigate transfer learning, the performance on languages that the model was not trained for does suffer.
Because this model used xlm-roberta-base as its starting point (potentially with domain adaptive fine-tuning on specific languages), this model's limitations can also apply here. These can include being biased towards the hegemonic viewpoint of most of its training data, being ungrounded and having subpar results on other languages (possibly due to unbalanced training data).
As [Adelani et al. (2021)](https://arxiv.org/abs/2103.11811) showed, the models in general struggled with entities that were either longer than 3 words and entities that were not contained in the training data. This could bias the models towards not finding, e.g. names of people that have many words, possibly leading to a misrepresentation in the results. Similarly, names that are uncommon, and may not have been found in the training data (due to e.g. different languages) would also be predicted less often.
Additionally, this model has not been verified in practice, and other, more subtle problems may become prevalent if used without any verification that it does what it is supposed to.
### Privacy & Ethical Considerations
The data comes from only publicly available news sources, the only available data should cover public figures and those that agreed to be reported on. See the original MasakhaNER paper for more details.
No explicit ethical considerations or adjustments were made during fine-tuning of this model.
## Metrics
The language adaptive models achieve (mostly) superior performance over starting with xlm-roberta-base. Our main metric was the aggregate F1 score for all NER categories.
These metrics are on the test set for MasakhaNER, so the data distribution is similar to the training set, so these results do not directly indicate how well these models generalise.
We do find large variation in transfer results when starting from different seeds (5 different seeds were tested), indicating that the fine-tuning process for transfer might be unstable.
The metrics used were chosen to be consistent with previous work, and to facilitate research. Other metrics may be more appropriate for other purposes.
## Caveats and Recommendations
In general, this model performed worse on the 'date' category compared to others, so if dates are a critical factor, then that might need to be taken into account and addressed, by for example collecting and annotating more data.
## Model Structure
Here are some performance details on this specific model, compared to others we trained.
All of these metrics were calculated on the test set, and the seed was chosen that gave the best overall F1 score. The first three result columns are averaged over all categories, and the latter 4 provide performance broken down by category.
This model can predict the following label for a token ([source](https://huggingface.co/Davlan/xlm-roberta-large-masakhaner)):
Abbreviation|Description
-|-
O|Outside of a named entity
B-DATE |Beginning of a DATE entity right after another DATE entity
I-DATE |DATE entity
B-PER |Beginning of a person’s name right after another person’s name
I-PER |Person’s name
B-ORG |Beginning of an organisation right after another organisation
I-ORG |Organisation
B-LOC |Beginning of a location right after another location
I-LOC |Location
| Model Name | Staring point | Evaluation / Fine-tune Language | F1 | Precision | Recall | F1 (DATE) | F1 (LOC) | F1 (ORG) | F1 (PER) |
| -------------------------------------------------- | -------------------- | -------------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- |
| [xlm-roberta-base-finetuned-igbo-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-igbo-finetuned-ner-swahili) (This model) | [ibo](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-igbo) | swa | 87.75 | 86.55 | 88.97 | 85.00 | 92.00 | 77.00 | 91.00 |
| [xlm-roberta-base-finetuned-hausa-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-hausa-finetuned-ner-swahili) | [hau](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-hausa) | swa | 88.36 | 86.95 | 89.82 | 86.00 | 91.00 | 77.00 | 94.00 |
| [xlm-roberta-base-finetuned-kinyarwanda-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-kinyarwanda-finetuned-ner-swahili) | [kin](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-kinyarwanda) | swa | 87.26 | 85.15 | 89.48 | 83.00 | 91.00 | 75.00 | 93.00 |
| [xlm-roberta-base-finetuned-luganda-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-luganda-finetuned-ner-swahili) | [lug](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-luganda) | swa | 88.93 | 87.64 | 90.25 | 83.00 | 92.00 | 79.00 | 95.00 |
| [xlm-roberta-base-finetuned-luo-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-luo-finetuned-ner-swahili) | [luo](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-luo) | swa | 87.93 | 86.91 | 88.97 | 83.00 | 91.00 | 76.00 | 94.00 |
| [xlm-roberta-base-finetuned-naija-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-naija-finetuned-ner-swahili) | [pcm](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-naija) | swa | 87.26 | 85.15 | 89.48 | 83.00 | 91.00 | 75.00 | 93.00 |
| [xlm-roberta-base-finetuned-swahili-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-swahili) | [swa](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-swahili) | swa | 90.36 | 88.59 | 92.20 | 86.00 | 93.00 | 79.00 | 96.00 |
| [xlm-roberta-base-finetuned-wolof-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-wolof-finetuned-ner-swahili) | [wol](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-wolof) | swa | 87.80 | 86.50 | 89.14 | 86.00 | 90.00 | 78.00 | 93.00 |
| [xlm-roberta-base-finetuned-yoruba-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-yoruba-finetuned-ner-swahili) | [yor](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-yoruba) | swa | 87.73 | 86.67 | 88.80 | 85.00 | 91.00 | 75.00 | 93.00 |
| [xlm-roberta-base-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-ner-swahili) | [base](https://huggingface.co/xlm-roberta-base) | swa | 88.71 | 86.84 | 90.67 | 83.00 | 91.00 | 79.00 | 95.00 |
## Usage
To use this model (or others), you can do the following, just changing the model name ([source](https://huggingface.co/dslim/bert-base-NER)):
```
from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
model_name = 'mbeukman/xlm-roberta-base-finetuned-igbo-finetuned-ner-swahili'
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
example = "Wizara ya afya ya Tanzania imeripoti Jumatatu kuwa , watu takriban 14 zaidi wamepata maambukizi ya Covid - 19 ."
ner_results = nlp(example)
print(ner_results)
```
|
mbeukman/xlm-roberta-base-finetuned-igbo-finetuned-ner-igbo | mbeukman | 2021-11-25T09:04:00Z | 4 | 0 | transformers | [
"transformers",
"pytorch",
"xlm-roberta",
"token-classification",
"NER",
"ig",
"dataset:masakhaner",
"arxiv:2103.11811",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:05Z | ---
language:
- ig
tags:
- NER
datasets:
- masakhaner
metrics:
- f1
- precision
- recall
widget:
- text: "Ike ịda jụụ otụ nkeji banyere oke ogbugbu na - eme n'ala Naijiria agwụla Ekweremmadụ"
---
# xlm-roberta-base-finetuned-igbo-finetuned-ner-igbo
This is a token classification (specifically NER) model that fine-tuned [xlm-roberta-base-finetuned-igbo](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-igbo) on the [MasakhaNER](https://arxiv.org/abs/2103.11811) dataset, specifically the Igbo part.
More information, and other similar models can be found in the [main Github repository](https://github.com/Michael-Beukman/NERTransfer).
## About
This model is transformer based and was fine-tuned on the MasakhaNER dataset. It is a named entity recognition dataset, containing mostly news articles in 10 different African languages.
The model was fine-tuned for 50 epochs, with a maximum sequence length of 200, 32 batch size, 5e-5 learning rate. This process was repeated 5 times (with different random seeds), and this uploaded model performed the best out of those 5 seeds (aggregate F1 on test set).
This model was fine-tuned by me, Michael Beukman while doing a project at the University of the Witwatersrand, Johannesburg. This is version 1, as of 20 November 2021.
This model is licensed under the [Apache License, Version 2.0](https://www.apache.org/licenses/LICENSE-2.0).
### Contact & More information
For more information about the models, including training scripts, detailed results and further resources, you can visit the the [main Github repository](https://github.com/Michael-Beukman/NERTransfer). You can contact me by filing an issue on this repository.
### Training Resources
In the interest of openness, and reporting resources used, we list here how long the training process took, as well as what the minimum resources would be to reproduce this. Fine-tuning each model on the NER dataset took between 10 and 30 minutes, and was performed on a NVIDIA RTX3090 GPU. To use a batch size of 32, at least 14GB of GPU memory was required, although it was just possible to fit these models in around 6.5GB's of VRAM when using a batch size of 1.
## Data
The train, evaluation and test datasets were taken directly from the MasakhaNER [Github](https://github.com/masakhane-io/masakhane-ner) repository, with minimal to no preprocessing, as the original dataset is already of high quality.
The motivation for the use of this data is that it is the "first large, publicly available, high quality dataset for named entity recognition (NER) in ten African languages" ([source](https://arxiv.org/pdf/2103.11811.pdf)). The high-quality data, as well as the groundwork laid by the paper introducing it are some more reasons why this dataset was used. For evaluation, the dedicated test split was used, which is from the same distribution as the training data, so this model may not generalise to other distributions, and further testing would need to be done to investigate this. The exact distribution of the data is covered in detail [here](https://arxiv.org/abs/2103.11811).
## Intended Use
This model are intended to be used for NLP research into e.g. interpretability or transfer learning. Using this model in production is not supported, as generalisability and downright performance is limited. In particular, this is not designed to be used in any important downstream task that could affect people, as harm could be caused by the limitations of the model, described next.
## Limitations
This model was only trained on one (relatively small) dataset, covering one task (NER) in one domain (news articles) and in a set span of time. The results may not generalise, and the model may perform badly, or in an unfair / biased way if used on other tasks. Although the purpose of this project was to investigate transfer learning, the performance on languages that the model was not trained for does suffer.
Because this model used xlm-roberta-base as its starting point (potentially with domain adaptive fine-tuning on specific languages), this model's limitations can also apply here. These can include being biased towards the hegemonic viewpoint of most of its training data, being ungrounded and having subpar results on other languages (possibly due to unbalanced training data).
As [Adelani et al. (2021)](https://arxiv.org/abs/2103.11811) showed, the models in general struggled with entities that were either longer than 3 words and entities that were not contained in the training data. This could bias the models towards not finding, e.g. names of people that have many words, possibly leading to a misrepresentation in the results. Similarly, names that are uncommon, and may not have been found in the training data (due to e.g. different languages) would also be predicted less often.
Additionally, this model has not been verified in practice, and other, more subtle problems may become prevalent if used without any verification that it does what it is supposed to.
### Privacy & Ethical Considerations
The data comes from only publicly available news sources, the only available data should cover public figures and those that agreed to be reported on. See the original MasakhaNER paper for more details.
No explicit ethical considerations or adjustments were made during fine-tuning of this model.
## Metrics
The language adaptive models achieve (mostly) superior performance over starting with xlm-roberta-base. Our main metric was the aggregate F1 score for all NER categories.
These metrics are on the test set for MasakhaNER, so the data distribution is similar to the training set, so these results do not directly indicate how well these models generalise.
We do find large variation in transfer results when starting from different seeds (5 different seeds were tested), indicating that the fine-tuning process for transfer might be unstable.
The metrics used were chosen to be consistent with previous work, and to facilitate research. Other metrics may be more appropriate for other purposes.
## Caveats and Recommendations
In general, this model performed worse on the 'date' category compared to others, so if dates are a critical factor, then that might need to be taken into account and addressed, by for example collecting and annotating more data.
## Model Structure
Here are some performance details on this specific model, compared to others we trained.
All of these metrics were calculated on the test set, and the seed was chosen that gave the best overall F1 score. The first three result columns are averaged over all categories, and the latter 4 provide performance broken down by category.
This model can predict the following label for a token ([source](https://huggingface.co/Davlan/xlm-roberta-large-masakhaner)):
Abbreviation|Description
-|-
O|Outside of a named entity
B-DATE |Beginning of a DATE entity right after another DATE entity
I-DATE |DATE entity
B-PER |Beginning of a person’s name right after another person’s name
I-PER |Person’s name
B-ORG |Beginning of an organisation right after another organisation
I-ORG |Organisation
B-LOC |Beginning of a location right after another location
I-LOC |Location
| Model Name | Staring point | Evaluation / Fine-tune Language | F1 | Precision | Recall | F1 (DATE) | F1 (LOC) | F1 (ORG) | F1 (PER) |
| -------------------------------------------------- | -------------------- | -------------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- |
| [xlm-roberta-base-finetuned-igbo-finetuned-ner-igbo](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-igbo-finetuned-ner-igbo) (This model) | [ibo](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-igbo) | ibo | 88.39 | 87.08 | 89.74 | 74.00 | 91.00 | 90.00 | 91.00 |
| [xlm-roberta-base-finetuned-swahili-finetuned-ner-igbo](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-igbo) | [swa](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-swahili) | ibo | 84.93 | 83.63 | 86.26 | 70.00 | 88.00 | 89.00 | 84.00 |
| [xlm-roberta-base-finetuned-ner-igbo](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-ner-igbo) | [base](https://huggingface.co/xlm-roberta-base) | ibo | 86.06 | 85.20 | 86.94 | 76.00 | 86.00 | 90.00 | 87.00 |
## Usage
To use this model (or others), you can do the following, just changing the model name ([source](https://huggingface.co/dslim/bert-base-NER)):
```
from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
model_name = 'mbeukman/xlm-roberta-base-finetuned-igbo-finetuned-ner-igbo'
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
example = "Ike ịda jụụ otụ nkeji banyere oke ogbugbu na - eme n'ala Naijiria agwụla Ekweremmadụ"
ner_results = nlp(example)
print(ner_results)
```
|
mbeukman/xlm-roberta-base-finetuned-hausa-finetuned-ner-swahili | mbeukman | 2021-11-25T09:03:58Z | 6 | 0 | transformers | [
"transformers",
"pytorch",
"xlm-roberta",
"token-classification",
"NER",
"sw",
"dataset:masakhaner",
"arxiv:2103.11811",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:05Z | ---
language:
- sw
tags:
- NER
datasets:
- masakhaner
metrics:
- f1
- precision
- recall
widget:
- text: "Wizara ya afya ya Tanzania imeripoti Jumatatu kuwa , watu takriban 14 zaidi wamepata maambukizi ya Covid - 19 ."
---
# xlm-roberta-base-finetuned-hausa-finetuned-ner-swahili
This is a token classification (specifically NER) model that fine-tuned [xlm-roberta-base-finetuned-hausa](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-hausa) on the [MasakhaNER](https://arxiv.org/abs/2103.11811) dataset, specifically the Swahili part.
More information, and other similar models can be found in the [main Github repository](https://github.com/Michael-Beukman/NERTransfer).
## About
This model is transformer based and was fine-tuned on the MasakhaNER dataset. It is a named entity recognition dataset, containing mostly news articles in 10 different African languages.
The model was fine-tuned for 50 epochs, with a maximum sequence length of 200, 32 batch size, 5e-5 learning rate. This process was repeated 5 times (with different random seeds), and this uploaded model performed the best out of those 5 seeds (aggregate F1 on test set).
This model was fine-tuned by me, Michael Beukman while doing a project at the University of the Witwatersrand, Johannesburg. This is version 1, as of 20 November 2021.
This model is licensed under the [Apache License, Version 2.0](https://www.apache.org/licenses/LICENSE-2.0).
### Contact & More information
For more information about the models, including training scripts, detailed results and further resources, you can visit the the [main Github repository](https://github.com/Michael-Beukman/NERTransfer). You can contact me by filing an issue on this repository.
### Training Resources
In the interest of openness, and reporting resources used, we list here how long the training process took, as well as what the minimum resources would be to reproduce this. Fine-tuning each model on the NER dataset took between 10 and 30 minutes, and was performed on a NVIDIA RTX3090 GPU. To use a batch size of 32, at least 14GB of GPU memory was required, although it was just possible to fit these models in around 6.5GB's of VRAM when using a batch size of 1.
## Data
The train, evaluation and test datasets were taken directly from the MasakhaNER [Github](https://github.com/masakhane-io/masakhane-ner) repository, with minimal to no preprocessing, as the original dataset is already of high quality.
The motivation for the use of this data is that it is the "first large, publicly available, high quality dataset for named entity recognition (NER) in ten African languages" ([source](https://arxiv.org/pdf/2103.11811.pdf)). The high-quality data, as well as the groundwork laid by the paper introducing it are some more reasons why this dataset was used. For evaluation, the dedicated test split was used, which is from the same distribution as the training data, so this model may not generalise to other distributions, and further testing would need to be done to investigate this. The exact distribution of the data is covered in detail [here](https://arxiv.org/abs/2103.11811).
## Intended Use
This model are intended to be used for NLP research into e.g. interpretability or transfer learning. Using this model in production is not supported, as generalisability and downright performance is limited. In particular, this is not designed to be used in any important downstream task that could affect people, as harm could be caused by the limitations of the model, described next.
## Limitations
This model was only trained on one (relatively small) dataset, covering one task (NER) in one domain (news articles) and in a set span of time. The results may not generalise, and the model may perform badly, or in an unfair / biased way if used on other tasks. Although the purpose of this project was to investigate transfer learning, the performance on languages that the model was not trained for does suffer.
Because this model used xlm-roberta-base as its starting point (potentially with domain adaptive fine-tuning on specific languages), this model's limitations can also apply here. These can include being biased towards the hegemonic viewpoint of most of its training data, being ungrounded and having subpar results on other languages (possibly due to unbalanced training data).
As [Adelani et al. (2021)](https://arxiv.org/abs/2103.11811) showed, the models in general struggled with entities that were either longer than 3 words and entities that were not contained in the training data. This could bias the models towards not finding, e.g. names of people that have many words, possibly leading to a misrepresentation in the results. Similarly, names that are uncommon, and may not have been found in the training data (due to e.g. different languages) would also be predicted less often.
Additionally, this model has not been verified in practice, and other, more subtle problems may become prevalent if used without any verification that it does what it is supposed to.
### Privacy & Ethical Considerations
The data comes from only publicly available news sources, the only available data should cover public figures and those that agreed to be reported on. See the original MasakhaNER paper for more details.
No explicit ethical considerations or adjustments were made during fine-tuning of this model.
## Metrics
The language adaptive models achieve (mostly) superior performance over starting with xlm-roberta-base. Our main metric was the aggregate F1 score for all NER categories.
These metrics are on the test set for MasakhaNER, so the data distribution is similar to the training set, so these results do not directly indicate how well these models generalise.
We do find large variation in transfer results when starting from different seeds (5 different seeds were tested), indicating that the fine-tuning process for transfer might be unstable.
The metrics used were chosen to be consistent with previous work, and to facilitate research. Other metrics may be more appropriate for other purposes.
## Caveats and Recommendations
In general, this model performed worse on the 'date' category compared to others, so if dates are a critical factor, then that might need to be taken into account and addressed, by for example collecting and annotating more data.
## Model Structure
Here are some performance details on this specific model, compared to others we trained.
All of these metrics were calculated on the test set, and the seed was chosen that gave the best overall F1 score. The first three result columns are averaged over all categories, and the latter 4 provide performance broken down by category.
This model can predict the following label for a token ([source](https://huggingface.co/Davlan/xlm-roberta-large-masakhaner)):
Abbreviation|Description
-|-
O|Outside of a named entity
B-DATE |Beginning of a DATE entity right after another DATE entity
I-DATE |DATE entity
B-PER |Beginning of a person’s name right after another person’s name
I-PER |Person’s name
B-ORG |Beginning of an organisation right after another organisation
I-ORG |Organisation
B-LOC |Beginning of a location right after another location
I-LOC |Location
| Model Name | Staring point | Evaluation / Fine-tune Language | F1 | Precision | Recall | F1 (DATE) | F1 (LOC) | F1 (ORG) | F1 (PER) |
| -------------------------------------------------- | -------------------- | -------------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- | -------------- |
| [xlm-roberta-base-finetuned-hausa-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-hausa-finetuned-ner-swahili) (This model) | [hau](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-hausa) | swa | 88.36 | 86.95 | 89.82 | 86.00 | 91.00 | 77.00 | 94.00 |
| [xlm-roberta-base-finetuned-igbo-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-igbo-finetuned-ner-swahili) | [ibo](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-igbo) | swa | 87.75 | 86.55 | 88.97 | 85.00 | 92.00 | 77.00 | 91.00 |
| [xlm-roberta-base-finetuned-kinyarwanda-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-kinyarwanda-finetuned-ner-swahili) | [kin](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-kinyarwanda) | swa | 87.26 | 85.15 | 89.48 | 83.00 | 91.00 | 75.00 | 93.00 |
| [xlm-roberta-base-finetuned-luganda-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-luganda-finetuned-ner-swahili) | [lug](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-luganda) | swa | 88.93 | 87.64 | 90.25 | 83.00 | 92.00 | 79.00 | 95.00 |
| [xlm-roberta-base-finetuned-luo-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-luo-finetuned-ner-swahili) | [luo](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-luo) | swa | 87.93 | 86.91 | 88.97 | 83.00 | 91.00 | 76.00 | 94.00 |
| [xlm-roberta-base-finetuned-naija-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-naija-finetuned-ner-swahili) | [pcm](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-naija) | swa | 87.26 | 85.15 | 89.48 | 83.00 | 91.00 | 75.00 | 93.00 |
| [xlm-roberta-base-finetuned-swahili-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-swahili) | [swa](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-swahili) | swa | 90.36 | 88.59 | 92.20 | 86.00 | 93.00 | 79.00 | 96.00 |
| [xlm-roberta-base-finetuned-wolof-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-wolof-finetuned-ner-swahili) | [wol](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-wolof) | swa | 87.80 | 86.50 | 89.14 | 86.00 | 90.00 | 78.00 | 93.00 |
| [xlm-roberta-base-finetuned-yoruba-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-yoruba-finetuned-ner-swahili) | [yor](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-yoruba) | swa | 87.73 | 86.67 | 88.80 | 85.00 | 91.00 | 75.00 | 93.00 |
| [xlm-roberta-base-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-ner-swahili) | [base](https://huggingface.co/xlm-roberta-base) | swa | 88.71 | 86.84 | 90.67 | 83.00 | 91.00 | 79.00 | 95.00 |
## Usage
To use this model (or others), you can do the following, just changing the model name ([source](https://huggingface.co/dslim/bert-base-NER)):
```
from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
model_name = 'mbeukman/xlm-roberta-base-finetuned-hausa-finetuned-ner-swahili'
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
example = "Wizara ya afya ya Tanzania imeripoti Jumatatu kuwa , watu takriban 14 zaidi wamepata maambukizi ya Covid - 19 ."
ner_results = nlp(example)
print(ner_results)
```
|
arnolfokam/mbert-base-uncased-pcm | arnolfokam | 2021-11-24T21:17:52Z | 8 | 0 | transformers | [
"transformers",
"pytorch",
"bert",
"token-classification",
"NER",
"pcm",
"dataset:masakhaner",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:05Z | ---
language:
- pcm
tags:
- NER
datasets:
- masakhaner
metrics:
- f1
- precision
- recall
license: apache-2.0
widget:
- text: "Mixed Martial Arts joinbodi, Ultimate Fighting Championship, UFC don decide say dem go enta back di octagon on Saturday, 9 May, for Jacksonville, Florida."
---
# Model description
**mbert-base-uncased-pcm** is a model based on the fine-tuned Multilingual BERT base uncased model. It has been trained to recognize four types of entities:
- dates & time (DATE)
- Location (LOC)
- Organizations (ORG)
- Person (PER)
# Intended Use
- Intended to be used for research purposes concerning Named Entity Recognition for African Languages.
- Not intended for practical purposes.
# Training Data
This model was fine-tuned on the Nigerian Pidgin corpus **(pcm)** of the [MasakhaNER](https://github.com/masakhane-io/masakhane-ner) dataset. However, we thresholded the number of entity groups per sentence in this dataset to 10 entity groups.
# Training procedure
This model was trained on a single NVIDIA P5000 from [Paperspace](https://www.paperspace.com)
#### Hyperparameters
- **Learning Rate:** 5e-5
- **Batch Size:** 32
- **Maximum Sequence Length:** 164
- **Epochs:** 30
# Evaluation Data
We evaluated this model on the test split of the Swahili corpus **(pcm)** present in the [MasakhaNER](https://github.com/masakhane-io/masakhane-ner) with no thresholding.
# Metrics
- Precision
- Recall
- F1-score
# Limitations
- The size of the pre-trained language model prevents its usage in anything other than research.
- Lack of analysis concerning the bias and fairness in these models may make them dangerous if deployed into production system.
- The train data is a less populated version of the original dataset in terms of entity groups per sentence. Therefore, this can negatively impact the performance.
# Caveats and Recommendations
- The topics in the dataset corpus are centered around **News**. Future training could be done with a more diverse corpus.
# Results
Model Name| Precision | Recall | F1-score
-|-|-|-
**mbert-base-uncased-pcm**| 90.46 | 83.23 | 86.69
# Usage
```python
from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
tokenizer = AutoTokenizer.from_pretrained("arnolfokam/mbert-base-uncased-pcm")
model = AutoModelForTokenClassification.from_pretrained("arnolfokam/mbert-base-uncased-pcm")
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
example = "Mixed Martial Arts joinbodi, Ultimate Fighting Championship, UFC don decide say dem go enta back di octagon on Saturday, 9 May, for Jacksonville, Florida."
ner_results = nlp(example)
print(ner_results)
``` |
bgoel4132/twitter-sentiment | bgoel4132 | 2021-11-24T19:39:02Z | 6 | 0 | transformers | [
"transformers",
"pytorch",
"bert",
"text-classification",
"autonlp",
"en",
"dataset:bgoel4132/autonlp-data-twitter-sentiment",
"co2_eq_emissions",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text-classification | 2022-03-02T23:29:05Z | ---
tags: autonlp
language: en
widget:
- text: "I love AutoNLP 🤗"
datasets:
- bgoel4132/autonlp-data-twitter-sentiment
co2_eq_emissions: 186.8637425115097
---
# Model Trained Using AutoNLP
- Problem type: Multi-class Classification
- Model ID: 35868888
- CO2 Emissions (in grams): 186.8637425115097
## Validation Metrics
- Loss: 0.2020547091960907
- Accuracy: 0.9233253193796257
- Macro F1: 0.9240407542958707
- Micro F1: 0.9233253193796257
- Weighted F1: 0.921800586774046
- Macro Precision: 0.9432284179846658
- Micro Precision: 0.9233253193796257
- Weighted Precision: 0.9247263361914827
- Macro Recall: 0.9139437626409382
- Micro Recall: 0.9233253193796257
- Weighted Recall: 0.9233253193796257
## Usage
You can use cURL to access this model:
```
$ curl -X POST -H "Authorization: Bearer YOUR_API_KEY" -H "Content-Type: application/json" -d '{"inputs": "I love AutoNLP"}' https://api-inference.huggingface.co/models/bgoel4132/autonlp-twitter-sentiment-35868888
```
Or Python API:
```
from transformers import AutoModelForSequenceClassification, AutoTokenizer
model = AutoModelForSequenceClassification.from_pretrained("bgoel4132/autonlp-twitter-sentiment-35868888", use_auth_token=True)
tokenizer = AutoTokenizer.from_pretrained("bgoel4132/autonlp-twitter-sentiment-35868888", use_auth_token=True)
inputs = tokenizer("I love AutoNLP", return_tensors="pt")
outputs = model(**inputs)
``` |
castorini/monot5-base-msmarco | castorini | 2021-11-24T17:59:19Z | 23,090 | 10 | transformers | [
"transformers",
"pytorch",
"jax",
"t5",
"text2text-generation",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text2text-generation | 2022-03-02T23:29:05Z | This model is a T5-base reranker fine-tuned on the MS MARCO passage dataset for 100k steps (or 10 epochs).
For better zero-shot performance (i.e., inference on other datasets), we recommend using `castorini/monot5-base-msmarco-10k`.
For more details on how to use it, check the following links:
- [A simple reranking example](https://github.com/castorini/pygaggle#a-simple-reranking-example)
- [Rerank MS MARCO passages](https://github.com/castorini/pygaggle/blob/master/docs/experiments-msmarco-passage-subset.md)
- [Rerank Robust04 documents](https://github.com/castorini/pygaggle/blob/master/docs/experiments-robust04-monot5-gpu.md)
Paper describing the model: [Document Ranking with a Pretrained Sequence-to-Sequence Model](https://www.aclweb.org/anthology/2020.findings-emnlp.63/) |
AdapterHub/roberta-base-pf-ud_pos | AdapterHub | 2021-11-24T16:32:48Z | 4 | 0 | adapter-transformers | [
"adapter-transformers",
"token-classification",
"roberta",
"adapterhub:pos/ud_ewt",
"en",
"dataset:universal_dependencies",
"arxiv:2104.08247",
"region:us"
] | token-classification | 2022-03-02T23:29:04Z | ---
tags:
- token-classification
- roberta
- adapterhub:pos/ud_ewt
- adapter-transformers
datasets:
- universal_dependencies
language:
- en
---
# Adapter `AdapterHub/roberta-base-pf-ud_pos` for roberta-base
An [adapter](https://adapterhub.ml) for the `roberta-base` model that was trained on the [pos/ud_ewt](https://adapterhub.ml/explore/pos/ud_ewt/) dataset and includes a prediction head for tagging.
This adapter was created for usage with the **[adapter-transformers](https://github.com/Adapter-Hub/adapter-transformers)** library.
## Usage
First, install `adapter-transformers`:
```
pip install -U adapter-transformers
```
_Note: adapter-transformers is a fork of transformers that acts as a drop-in replacement with adapter support. [More](https://docs.adapterhub.ml/installation.html)_
Now, the adapter can be loaded and activated like this:
```python
from transformers import AutoModelWithHeads
model = AutoModelWithHeads.from_pretrained("roberta-base")
adapter_name = model.load_adapter("AdapterHub/roberta-base-pf-ud_pos", source="hf")
model.active_adapters = adapter_name
```
## Architecture & Training
The training code for this adapter is available at https://github.com/adapter-hub/efficient-task-transfer.
In particular, training configurations for all tasks can be found [here](https://github.com/adapter-hub/efficient-task-transfer/tree/master/run_configs).
## Evaluation results
Refer to [the paper](https://arxiv.org/pdf/2104.08247) for more information on results.
## Citation
If you use this adapter, please cite our paper ["What to Pre-Train on? Efficient Intermediate Task Selection"](https://arxiv.org/pdf/2104.08247):
```bibtex
@inproceedings{poth-etal-2021-pre,
title = "{W}hat to Pre-Train on? {E}fficient Intermediate Task Selection",
author = {Poth, Clifton and
Pfeiffer, Jonas and
R{"u}ckl{'e}, Andreas and
Gurevych, Iryna},
booktitle = "Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing",
month = nov,
year = "2021",
address = "Online and Punta Cana, Dominican Republic",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2021.emnlp-main.827",
pages = "10585--10605",
}
``` |
AdapterHub/roberta-base-pf-trec | AdapterHub | 2021-11-24T16:32:34Z | 0 | 0 | adapter-transformers | [
"adapter-transformers",
"text-classification",
"roberta",
"en",
"dataset:trec",
"arxiv:2104.08247",
"region:us"
] | text-classification | 2022-03-02T23:29:04Z | ---
tags:
- text-classification
- roberta
- adapter-transformers
datasets:
- trec
language:
- en
---
# Adapter `AdapterHub/roberta-base-pf-trec` for roberta-base
An [adapter](https://adapterhub.ml) for the `roberta-base` model that was trained on the [trec](https://huggingface.co/datasets/trec/) dataset and includes a prediction head for classification.
This adapter was created for usage with the **[adapter-transformers](https://github.com/Adapter-Hub/adapter-transformers)** library.
## Usage
First, install `adapter-transformers`:
```
pip install -U adapter-transformers
```
_Note: adapter-transformers is a fork of transformers that acts as a drop-in replacement with adapter support. [More](https://docs.adapterhub.ml/installation.html)_
Now, the adapter can be loaded and activated like this:
```python
from transformers import AutoModelWithHeads
model = AutoModelWithHeads.from_pretrained("roberta-base")
adapter_name = model.load_adapter("AdapterHub/roberta-base-pf-trec", source="hf")
model.active_adapters = adapter_name
```
## Architecture & Training
The training code for this adapter is available at https://github.com/adapter-hub/efficient-task-transfer.
In particular, training configurations for all tasks can be found [here](https://github.com/adapter-hub/efficient-task-transfer/tree/master/run_configs).
## Evaluation results
Refer to [the paper](https://arxiv.org/pdf/2104.08247) for more information on results.
## Citation
If you use this adapter, please cite our paper ["What to Pre-Train on? Efficient Intermediate Task Selection"](https://arxiv.org/pdf/2104.08247):
```bibtex
@inproceedings{poth-etal-2021-pre,
title = "{W}hat to Pre-Train on? {E}fficient Intermediate Task Selection",
author = {Poth, Clifton and
Pfeiffer, Jonas and
R{"u}ckl{'e}, Andreas and
Gurevych, Iryna},
booktitle = "Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing",
month = nov,
year = "2021",
address = "Online and Punta Cana, Dominican Republic",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2021.emnlp-main.827",
pages = "10585--10605",
}
``` |
AdapterHub/roberta-base-pf-swag | AdapterHub | 2021-11-24T16:32:26Z | 0 | 0 | adapter-transformers | [
"adapter-transformers",
"roberta",
"en",
"dataset:swag",
"arxiv:2104.08247",
"region:us"
] | null | 2022-03-02T23:29:04Z | ---
tags:
- roberta
- adapter-transformers
datasets:
- swag
language:
- en
---
# Adapter `AdapterHub/roberta-base-pf-swag` for roberta-base
An [adapter](https://adapterhub.ml) for the `roberta-base` model that was trained on the [swag](https://huggingface.co/datasets/swag/) dataset and includes a prediction head for multiple choice.
This adapter was created for usage with the **[adapter-transformers](https://github.com/Adapter-Hub/adapter-transformers)** library.
## Usage
First, install `adapter-transformers`:
```
pip install -U adapter-transformers
```
_Note: adapter-transformers is a fork of transformers that acts as a drop-in replacement with adapter support. [More](https://docs.adapterhub.ml/installation.html)_
Now, the adapter can be loaded and activated like this:
```python
from transformers import AutoModelWithHeads
model = AutoModelWithHeads.from_pretrained("roberta-base")
adapter_name = model.load_adapter("AdapterHub/roberta-base-pf-swag", source="hf")
model.active_adapters = adapter_name
```
## Architecture & Training
The training code for this adapter is available at https://github.com/adapter-hub/efficient-task-transfer.
In particular, training configurations for all tasks can be found [here](https://github.com/adapter-hub/efficient-task-transfer/tree/master/run_configs).
## Evaluation results
Refer to [the paper](https://arxiv.org/pdf/2104.08247) for more information on results.
## Citation
If you use this adapter, please cite our paper ["What to Pre-Train on? Efficient Intermediate Task Selection"](https://arxiv.org/pdf/2104.08247):
```bibtex
@inproceedings{poth-etal-2021-what-to-pre-train-on,
title={What to Pre-Train on? Efficient Intermediate Task Selection},
author={Clifton Poth and Jonas Pfeiffer and Andreas Rücklé and Iryna Gurevych},
booktitle = "Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (EMNLP)",
month = nov,
year = "2021",
address = "Online",
publisher = "Association for Computational Linguistics",
url = "https://arxiv.org/abs/2104.08247",
pages = "to appear",
}
``` |
AdapterHub/roberta-base-pf-social_i_qa | AdapterHub | 2021-11-24T16:32:05Z | 5 | 0 | adapter-transformers | [
"adapter-transformers",
"roberta",
"en",
"dataset:social_i_qa",
"arxiv:2104.08247",
"region:us"
] | null | 2022-03-02T23:29:04Z | ---
tags:
- roberta
- adapter-transformers
datasets:
- social_i_qa
language:
- en
---
# Adapter `AdapterHub/roberta-base-pf-social_i_qa` for roberta-base
An [adapter](https://adapterhub.ml) for the `roberta-base` model that was trained on the [social_i_qa](https://huggingface.co/datasets/social_i_qa/) dataset and includes a prediction head for multiple choice.
This adapter was created for usage with the **[adapter-transformers](https://github.com/Adapter-Hub/adapter-transformers)** library.
## Usage
First, install `adapter-transformers`:
```
pip install -U adapter-transformers
```
_Note: adapter-transformers is a fork of transformers that acts as a drop-in replacement with adapter support. [More](https://docs.adapterhub.ml/installation.html)_
Now, the adapter can be loaded and activated like this:
```python
from transformers import AutoModelWithHeads
model = AutoModelWithHeads.from_pretrained("roberta-base")
adapter_name = model.load_adapter("AdapterHub/roberta-base-pf-social_i_qa", source="hf")
model.active_adapters = adapter_name
```
## Architecture & Training
The training code for this adapter is available at https://github.com/adapter-hub/efficient-task-transfer.
In particular, training configurations for all tasks can be found [here](https://github.com/adapter-hub/efficient-task-transfer/tree/master/run_configs).
## Evaluation results
Refer to [the paper](https://arxiv.org/pdf/2104.08247) for more information on results.
## Citation
If you use this adapter, please cite our paper ["What to Pre-Train on? Efficient Intermediate Task Selection"](https://arxiv.org/pdf/2104.08247):
```bibtex
@inproceedings{poth-etal-2021-what-to-pre-train-on,
title={What to Pre-Train on? Efficient Intermediate Task Selection},
author={Clifton Poth and Jonas Pfeiffer and Andreas Rücklé and Iryna Gurevych},
booktitle = "Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (EMNLP)",
month = nov,
year = "2021",
address = "Online",
publisher = "Association for Computational Linguistics",
url = "https://arxiv.org/abs/2104.08247",
pages = "to appear",
}
``` |
AdapterHub/roberta-base-pf-sick | AdapterHub | 2021-11-24T16:31:49Z | 10 | 1 | adapter-transformers | [
"adapter-transformers",
"text-classification",
"roberta",
"adapterhub:nli/sick",
"en",
"dataset:sick",
"arxiv:2104.08247",
"region:us"
] | text-classification | 2022-03-02T23:29:04Z | ---
tags:
- text-classification
- roberta
- adapter-transformers
- adapterhub:nli/sick
- text-classification
datasets:
- sick
language:
- en
---
# Adapter `AdapterHub/roberta-base-pf-sick` for roberta-base
An [adapter](https://adapterhub.ml) for the `roberta-base` model that was trained on the [nli/sick](https://adapterhub.ml/explore/nli/sick/) dataset and includes a prediction head for classification.
This adapter was created for usage with the **[adapter-transformers](https://github.com/Adapter-Hub/adapter-transformers)** library.
## Usage
First, install `adapter-transformers`:
```
pip install -U adapter-transformers
```
_Note: adapter-transformers is a fork of transformers that acts as a drop-in replacement with adapter support. [More](https://docs.adapterhub.ml/installation.html)_
Now, the adapter can be loaded and activated like this:
```python
from transformers import AutoModelWithHeads
model = AutoModelWithHeads.from_pretrained("roberta-base")
adapter_name = model.load_adapter("AdapterHub/roberta-base-pf-sick", source="hf")
model.active_adapters = adapter_name
```
## Architecture & Training
The training code for this adapter is available at https://github.com/adapter-hub/efficient-task-transfer.
In particular, training configurations for all tasks can be found [here](https://github.com/adapter-hub/efficient-task-transfer/tree/master/run_configs).
## Evaluation results
Refer to [the paper](https://arxiv.org/pdf/2104.08247) for more information on results.
## Citation
If you use this adapter, please cite our paper ["What to Pre-Train on? Efficient Intermediate Task Selection"](https://arxiv.org/pdf/2104.08247):
```bibtex
@inproceedings{poth-etal-2021-pre,
title = "{W}hat to Pre-Train on? {E}fficient Intermediate Task Selection",
author = {Poth, Clifton and
Pfeiffer, Jonas and
R{"u}ckl{'e}, Andreas and
Gurevych, Iryna},
booktitle = "Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing",
month = nov,
year = "2021",
address = "Online and Punta Cana, Dominican Republic",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2021.emnlp-main.827",
pages = "10585--10605",
}
``` |
Subsets and Splits