modelId
stringlengths 5
139
| author
stringlengths 2
42
| last_modified
timestamp[us, tz=UTC]date 2020-02-15 11:33:14
2025-06-23 12:29:03
| downloads
int64 0
223M
| likes
int64 0
11.7k
| library_name
stringclasses 492
values | tags
sequencelengths 1
4.05k
| pipeline_tag
stringclasses 54
values | createdAt
timestamp[us, tz=UTC]date 2022-03-02 23:29:04
2025-06-23 12:24:08
| card
stringlengths 11
1.01M
|
|---|---|---|---|---|---|---|---|---|---|
pere/nb-roberta-base-scandinavian-long | pere | 2021-11-25T18:21:53Z | 5 | 0 | transformers | [
"transformers",
"pytorch",
"jax",
"tensorboard",
"roberta",
"fill-mask",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | fill-mask | 2022-03-02T23:29:05Z | # This is just a Test Model. Do NOT use for anything!
Continued pretrained from the nb-roberta-base.
The domain specific pretraining is done on the 102GB (Scandinavian corpus)[https://huggingface.co/datasets/NbAiLab/scandinavian].
## Train for 180k steps for 128 sequences:
```bash
./run_mlm_flax_stream.py \
--output_dir="./" \
--model_type="roberta" \
--config_name="./" \
--tokenizer_name="./" \
--model_name_or_path="./" \
--dataset_name="NbAiLab/scandinavian" \
--max_seq_length="128" \
--weight_decay="0.01" \
--per_device_train_batch_size="128" \
--per_device_eval_batch_size="128" \
--learning_rate="6e-5" \
--warmup_steps="5000" \
--overwrite_output_dir \
--cache_dir /mnt/disks/flaxdisk/cache/ \
--num_train_steps="180000" \
--adam_beta1="0.9" \
--adam_beta2="0.98" \
--logging_steps="10000" \
--save_steps="10000" \
--eval_steps="10000" \
--preprocessing_num_workers 96 \
--auth_token True \
--adafactor \
--push_to_hub
```
## Train for 20k steps for 512 sequences:
```bash
./run_mlm_flax_stream.py \
--output_dir="./" \
--model_type="roberta" \
--config_name="./" \
--tokenizer_name="./" \
--model_name_or_path="./" \
--dataset_name="NbAiLab/scandinavian" \
--max_seq_length="512" \
--weight_decay="0.01" \
--per_device_train_batch_size="48" \
--per_device_eval_batch_size="48" \
--learning_rate="3e-5" \
--warmup_steps="5000" \
--overwrite_output_dir \
--cache_dir /mnt/disks/flaxdisk/cache/ \
--num_train_steps="20000" \
--adam_beta1="0.9" \
--adam_beta2="0.98" \
--logging_steps="20000" \
--save_steps="10000" \
--eval_steps="10000" \
--preprocessing_num_workers 96 \
--auth_token True \
--adafactor \
--push_to_hub
```
Approximate additional training time: 1 week.
|
pere/norwegian-roberta-base-highlr-512 | pere | 2021-11-25T17:54:31Z | 3 | 0 | transformers | [
"transformers",
"pytorch",
"jax",
"tensorboard",
"roberta",
"fill-mask",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | fill-mask | 2022-03-02T23:29:05Z | Same as norwegian-roberta-base but with higher learning rate and batch size |
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">๐ค AI BOT ๐ค</div>
<div style="text-align: center; font-size: 16px; font-weight: 800">รzgรผr Demirtaล</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 รzgรผr Demirtaล.
| Data | รzgรผr Demirtaล |
| --- | --- |
| 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)
|
espnet/kan-bayashi_csj_asr_train_asr_conformer | espnet | 2021-11-25T09:30:10Z | 5 | 1 | espnet | [
"espnet",
"audio",
"automatic-speech-recognition",
"jp",
"dataset:csj",
"arxiv:1804.00015",
"license:cc-by-4.0",
"region:us"
] | automatic-speech-recognition | 2022-03-02T23:29:05Z | ---
tags:
- espnet
- audio
- automatic-speech-recognition
language: jp
datasets:
- csj
license: cc-by-4.0
---
## ESPnet2 ASR model
### `espnet/kan-bayashi_csj_asr_train_asr_conformer`
This model was trained by Nelson Yalta using csj recipe in [espnet](https://github.com/espnet/espnet/).
### Demo: How to use in ESPnet2
```bash
cd espnet
git checkout 0d8cd47dd3572248b502bc831cd305e648170233
pip install -e .
cd egs2/csj/asr1
./run.sh --skip_data_prep false --skip_train true --download_model espnet/kan-bayashi_csj_asr_train_asr_conformer
```
## ASR config
<details><summary>expand</summary>
```
config: conf/tuning/train_asr_conformer.yaml
print_config: false
log_level: INFO
dry_run: false
iterator_type: sequence
output_dir: exp/asr_train_asr_conformer_raw_char_sp
ngpu: 1
seed: 0
num_workers: 1
num_att_plot: 3
dist_backend: nccl
dist_init_method: env://
dist_world_size: 4
dist_rank: 0
local_rank: 0
dist_master_addr: localhost
dist_master_port: 47308
dist_launcher: null
multiprocessing_distributed: true
cudnn_enabled: true
cudnn_benchmark: false
cudnn_deterministic: true
collect_stats: false
write_collected_feats: false
max_epoch: 100
patience: null
val_scheduler_criterion:
- valid
- loss
early_stopping_criterion:
- valid
- loss
- min
best_model_criterion:
- - valid
- acc
- max
keep_nbest_models: 10
grad_clip: 5.0
grad_clip_type: 2.0
grad_noise: false
accum_grad: 6
no_forward_run: false
resume: true
train_dtype: float32
use_amp: false
log_interval: null
pretrain_path: []
pretrain_key: []
num_iters_per_epoch: null
batch_size: 20
valid_batch_size: null
batch_bins: 15000000
valid_batch_bins: null
train_shape_file:
- exp/asr_stats_raw_sp/train/speech_shape
- exp/asr_stats_raw_sp/train/text_shape.char
valid_shape_file:
- exp/asr_stats_raw_sp/valid/speech_shape
- exp/asr_stats_raw_sp/valid/text_shape.char
batch_type: numel
valid_batch_type: null
fold_length:
- 80000
- 150
sort_in_batch: descending
sort_batch: descending
multiple_iterator: false
chunk_length: 500
chunk_shift_ratio: 0.5
num_cache_chunks: 1024
train_data_path_and_name_and_type:
- - dump/raw/train_nodup_sp/wav.scp
- speech
- sound
- - dump/raw/train_nodup_sp/text
- text
- text
valid_data_path_and_name_and_type:
- - dump/raw/train_dev/wav.scp
- speech
- sound
- - dump/raw/train_dev/text
- text
- text
allow_variable_data_keys: false
max_cache_size: 0.0
valid_max_cache_size: null
optim: adam
optim_conf:
lr: 0.002
scheduler: warmuplr
scheduler_conf:
warmup_steps: 25000
token_list:
- <blank>
- <unk>
- "\u306E"
- "\u3044"
- "\u3067"
- "\u3068"
- "\u30FC"
- "\u3066"
- "\u3046"
- "\u307E"
- "\u3059"
- "\u3057"
- "\u306B"
- "\u3063"
- "\u306A"
- "\u3048"
- "\u305F"
- "\u3053"
- "\u304C"
- "\u304B"
- "\u306F"
- "\u308B"
- "\u3042"
- "\u3093"
- "\u308C"
- "\u3082"
- "\u3092"
- "\u305D"
- "\u308A"
- "\u3089"
- "\u3051"
- "\u304F"
- "\u3069"
- "\u3088"
- "\u304D"
- "\u3060"
- "\u304A"
- "\u30F3"
- "\u306D"
- "\u4E00"
- "\u3055"
- "\u30B9"
- "\u8A00"
- "\u3061"
- "\u3064"
- "\u5206"
- "\u30C8"
- "\u3084"
- "\u4EBA"
- "\u30EB"
- "\u601D"
- "\u308F"
- "\u6642"
- "\u65B9"
- "\u3058"
- "\u30A4"
- "\u884C"
- "\u4F55"
- "\u307F"
- "\u5341"
- "\u30E9"
- "\u4E8C"
- "\u672C"
- "\u8A9E"
- "\u5927"
- "\u7684"
- "\u30AF"
- "\u30BF"
- "\u308D"
- "\u3070"
- "\u3087"
- "\u3083"
- "\u97F3"
- "\u51FA"
- "\u305B"
- "\u30C3"
- "\u5408"
- "\u65E5"
- "\u4E2D"
- "\u751F"
- "\u4ECA"
- "\u898B"
- "\u30EA"
- "\u9593"
- "\u8A71"
- "\u3081"
- "\u30A2"
- "\u5F8C"
- "\u81EA"
- "\u305A"
- "\u79C1"
- "\u30C6"
- "\u4E0A"
- "\u5E74"
- "\u5B66"
- "\u4E09"
- "\u30B7"
- "\u5834"
- "\u30C7"
- "\u5B9F"
- "\u5B50"
- "\u4F53"
- "\u8003"
- "\u5BFE"
- "\u7528"
- "\u6587"
- "\u30D1"
- "\u5F53"
- "\u7D50"
- "\u5EA6"
- "\u5165"
- "\u8A33"
- "\u30D5"
- "\u98A8"
- "\u30E0"
- "\u30D7"
- "\u6700"
- "\u30C9"
- "\u30EC"
- "\u30ED"
- "\u4F5C"
- "\u6570"
- "\u76EE"
- "\u30B8"
- "\u95A2"
- "\u30B0"
- "\u767A"
- "\u8005"
- "\u5B9A"
- "\u3005"
- "\u3050"
- "\u30B3"
- "\u4E8B"
- "\u624B"
- "\u5168"
- "\u5909"
- "\u30DE"
- "\u6027"
- "\u8868"
- "\u4F8B"
- "\u52D5"
- "\u8981"
- "\u5148"
- "\u524D"
- "\u610F"
- "\u90E8"
- "\u4F1A"
- "\u6301"
- "\u30E1"
- "\u5316"
- "\u9054"
- "\u4ED8"
- "\u5F62"
- "\u73FE"
- "\u4E94"
- "\u30AB"
- "\u3079"
- "\u53D6"
- "\u56DE"
- "\u5E38"
- "\u4F7F"
- "\u611F"
- "\u66F8"
- "\u6C17"
- "\u6CD5"
- "\u7A0B"
- "\u3071"
- "\u56DB"
- "\u591A"
- "\u8272"
- "\u30BB"
- "\u7406"
- "\u975E"
- "\u30D0"
- "\u58F0"
- "\u5358"
- "\u756A"
- "\uFF21"
- "\u6210"
- "\u540C"
- "\u901A"
- "\u30A3"
- "\u679C"
- "\u30AD"
- "\u554F"
- "\u984C"
- "\u69CB"
- "\u56FD"
- "\u6765"
- "\u9AD8"
- "\u6B21"
- "\u9A13"
- "\u3052"
- "\u30C1"
- "\u4EE5"
- "\u3054"
- "\u4EE3"
- "\u30E2"
- "\u30AA"
- "\u51C4"
- "\u7279"
- "\u77E5"
- "\u30E5"
- "\u7269"
- "\u660E"
- "\u70B9"
- "\u5473"
- "\u767E"
- "\u89E3"
- "\u8FD1"
- "\u8B58"
- "\u5730"
- "\u540D"
- "\u805E"
- "\u4E0B"
- "\u5C0F"
- "\u6559"
- "\u30B5"
- "\u70BA"
- "\u4E5D"
- "\u30D6"
- "\u5BB6"
- "\u30CB"
- "\u521D"
- "\u30D9"
- "\u30E7"
- "\u5C11"
- "\u8A8D"
- "\u8AD6"
- "\u529B"
- "\u516D"
- "\u30D3"
- "\u60C5"
- "\u7FD2"
- "\u30A6"
- "\u7ACB"
- "\u5FC3"
- "\u8ABF"
- "\u5831"
- "\u30A8"
- "\uFF24"
- "\uFF2E"
- "\u793A"
- "\u793E"
- "\u9055"
- "\u969B"
- "\u3056"
- "\u8AAC"
- "\u5FDC"
- "\u98DF"
- "\u72B6"
- "\u9577"
- "\u7814"
- "\u6821"
- "\u5185"
- "\u639B"
- "\u30DF"
- "\u5916"
- "\u5411"
- "\u80FD"
- "\u516B"
- "\u9762"
- "\u7A76"
- "\u7136"
- "\u3073"
- "\u30D4"
- "\u4E3B"
- "\u4FC2"
- "\u5024"
- "\u91CD"
- "\u8A5E"
- "\u4F9B"
- "\u5F97"
- "\u5FC5"
- "\u5973"
- "\u78BA"
- "\u7D42"
- "\u30BA"
- "\u6BCD"
- "\u696D"
- "\u7387"
- "\u65B0"
- "\u6D3B"
- "\u697D"
- "\u8449"
- "\u8A08"
- "\u30CA"
- "\u3080"
- "\u6240"
- "\u4E16"
- "\u6B63"
- "\u30E3"
- "\u8A18"
- "\u671F"
- "\u5207"
- "\u3078"
- "\u6A5F"
- "\u30DA"
- "\u5343"
- "\u985E"
- "\u5143"
- "\u614B"
- "\u826F"
- "\u5728"
- "\u6709"
- "\u30C0"
- "\u4E03"
- "\uFF23"
- "\u5225"
- "\u30EF"
- "\u691C"
- "\u7D9A"
- "\u9078"
- "\u57FA"
- "\u76F8"
- "\u6708"
- "\u4FA1"
- "\u7D20"
- "\u4ED6"
- "\u6BD4"
- "\u9023"
- "\u96C6"
- "\u30A7"
- "\u307B"
- "\u4F4D"
- "\u597D"
- "\uFF2D"
- "\u5F37"
- "\u4E0D"
- "\u5FA1"
- "\u6790"
- "\u30DD"
- "\u7121"
- "\u89AA"
- "\u53D7"
- "\u3086"
- "\u7F6E"
- "\u8C61"
- "\u4ED5"
- "\u5F0F"
- "\u30CD"
- "\u6307"
- "\u8AAD"
- "\u6C7A"
- "\u8ECA"
- "\u96FB"
- "\u904E"
- "\u30B1"
- "\u8A55"
- "\u5229"
- "\u6B8B"
- "\u8D77"
- "\u30CE"
- "\u7D4C"
- "\u56F3"
- "\u4F1D"
- "\u500B"
- "\u30C4"
- "\u7BC0"
- "\u9053"
- "\u5E73"
- "\u91D1"
- "\u899A"
- "\uFF34"
- "\u4F4F"
- "\u59CB"
- "\u63D0"
- "\u5B58"
- "\u5171"
- "\u30DB"
- "\u7B2C"
- "\u7D44"
- "\u89B3"
- "\u80B2"
- "\u6771"
- "\u305E"
- "\u958B"
- "\u52A0"
- "\u5F15"
- "\uFF33"
- "\u53E3"
- "\u6C34"
- "\u5BB9"
- "\u5468"
- "\u5B87"
- "\u7D04"
- "\u5B57"
- "\u3076"
- "\u9803"
- "\u3072"
- "\u5B99"
- "\u6BB5"
- "\u30BD"
- "\u97FF"
- "\u30DC"
- "\u53CB"
- "\u91CF"
- "\u6599"
- "\u3085"
- "\u5CF6"
- "\u8EAB"
- "\u76F4"
- "\u753B"
- "\u7DDA"
- "\u54C1"
- "\u5DEE"
- "\u4EF6"
- "\u9069"
- "\u5F35"
- "\u8FBA"
- "\u8FBC"
- "\u91CE"
- "\u69D8"
- "\u578B"
- "\u4E88"
- "\u7A2E"
- "\u5074"
- "\u8FF0"
- "\u5C71"
- "\u5C4B"
- "\u5E30"
- "\u30CF"
- "\u4E57"
- "\u539F"
- "\u683C"
- "\u8CEA"
- "\u666E"
- "\uFF30"
- "\u9020"
- "\u753A"
- "\u30B4"
- "\u82F1"
- "\u63A5"
- "\u304E"
- "\u6E2C"
- "\u3075"
- "\u7FA9"
- "\u4EAC"
- "\u5272"
- "\u5236"
- "\u7B54"
- "\u5404"
- "\u4FE1"
- "\u754C"
- "\u6211"
- "\u7A7A"
- "\uFF0E"
- "\u7740"
- "\u53EF"
- "\u66F4"
- "\u6D77"
- "\u4E0E"
- "\u9032"
- "\u52B9"
- "\u5F7C"
- "\u771F"
- "\u7530"
- "\u5FB4"
- "\u6D41"
- "\u5177"
- "\uFF32"
- "\u5E02"
- "\u67FB"
- "\u5B89"
- "\uFF22"
- "\u5E83"
- "\u50D5"
- "\u6CE2"
- "\u5C40"
- "\u8A2D"
- "\u7537"
- "\u767D"
- "\u30B6"
- "\u53CD"
- "\u6226"
- "\u533A"
- "\u6C42"
- "\u96D1"
- "\uFF29"
- "\u6B69"
- "\u8CB7"
- "\u982D"
- "\u7B97"
- "\u534A"
- "\u4FDD"
- "\u5E03"
- "\u96E3"
- "\uFF2C"
- "\u5224"
- "\u843D"
- "\u8DB3"
- "\u5E97"
- "\u7533"
- "\u8FD4"
- "\u30AE"
- "\u4E07"
- "\u6728"
- "\u6614"
- "\u8F03"
- "\u7D22"
- "\uFF26"
- "\u30B2"
- "\u6B86"
- "\u60AA"
- "\u5883"
- "\u548C"
- "\u907A"
- "\u57DF"
- "\u968E"
- "\u542B"
- "\u305C"
- "\u30BC"
- "\u65AD"
- "\u9650"
- "\u63A8"
- "\u4F4E"
- "\u5F71"
- "\u898F"
- "\u6319"
- "\u90FD"
- "\u307C"
- "\u6848"
- "\u4EEE"
- "\u88AB"
- "\u547C"
- "\u30A1"
- "\u96E2"
- "\u7CFB"
- "\u79FB"
- "\u30AC"
- "\u5DDD"
- "\u6E96"
- "\u904B"
- "\u6761"
- "\u5FF5"
- "\u6C11"
- "\uFF27"
- "\u7236"
- "\u75C5"
- "\u79D1"
- "\u4E21"
- "\u7531"
- "\u8A66"
- "\u56E0"
- "\u547D"
- "\u795E"
- "\uFF28"
- "\u7570"
- "\u7C21"
- "\u53E4"
- "\u6F14"
- "\u5897"
- "\u51E6"
- "\u8B70"
- "\u7DD2"
- "\u7CBE"
- "\u6613"
- "\u53F7"
- "\u65CF"
- "\u52FF"
- "\u60F3"
- "\u5217"
- "\u5C0E"
- "\u8EE2"
- "\u54E1"
- "\u30E6"
- "\u6BCE"
- "\u8996"
- "\u4E26"
- "\u98DB"
- "\u4F3C"
- "\u6620"
- "\u7D71"
- "\u4EA4"
- "\u30D2"
- "\u6B4C"
- "\u5F85"
- "\u8CC7"
- "\u8907"
- "\u8AA4"
- "\u63DB"
- "\u6A19"
- "\u6CC1"
- "\u914D"
- "\u62BD"
- "\u822C"
- "\u7403"
- "\u9006"
- "\u65C5"
- "\u6628"
- "\u9662"
- "\u99C5"
- "\u74B0"
- "\u5BDF"
- "\u516C"
- "\u6B73"
- "\u5C5E"
- "\u8F9E"
- "\u5947"
- "\u6CBB"
- "\u5E7E"
- "\u82E5"
- "\u58F2"
- "\u632F"
- "\u7686"
- "\u6CE8"
- "\u6B74"
- "\u9805"
- "\u5F93"
- "\u5747"
- "\u5F79"
- "\u9806"
- "\u53BB"
- "\u56E3"
- "\u8853"
- "\u7DF4"
- "\u6FC0"
- "\u6982"
- "\u66FF"
- "\u7B49"
- "\u98F2"
- "\u53F2"
- "\u88DC"
- "\u901F"
- "\u53C2"
- "\u65E9"
- "\u53CE"
- "\u9332"
- "\u671D"
- "\u5186"
- "\u5370"
- "\u5668"
- "\u63A2"
- "\u7D00"
- "\u9001"
- "\u6E1B"
- "\u571F"
- "\u5929"
- "\uFF2F"
- "\u50BE"
- "\u72AC"
- "\u9060"
- "\u5E2F"
- "\u52A9"
- "\u6A2A"
- "\u591C"
- "\u7523"
- "\u8AB2"
- "\u5BA2"
- "\u629E"
- "\u5712"
- "\u4E38"
- "\u50CF"
- "\u50CD"
- "\u6750"
- "\u5DE5"
- "\u904A"
- "\u544A"
- "\u523A"
- "\u6539"
- "\u8D64"
- "\u8074"
- "\u4ECB"
- "\u8077"
- "\u53F0"
- "\u77ED"
- "\u8AB0"
- "\u7D30"
- "\u672A"
- "\u770C"
- "\u9928"
- "\u6B62"
- "\u53F3"
- "\u306C"
- "\u3065"
- "\u56F2"
- "\u8A0E"
- "\u6B7B"
- "\u5EFA"
- "\u592B"
- "\u7AE0"
- "\u964D"
- "\u666F"
- "\u706B"
- "\u30A9"
- "\u9E97"
- "\u8B1B"
- "\u72EC"
- "\u5DE6"
- "\u5C64"
- "\uFF25"
- "\u5C55"
- "\u653F"
- "\u5099"
- "\u4F59"
- "\u7D76"
- "\u5065"
- "\u518D"
- "\u9580"
- "\u5546"
- "\u52DD"
- "\u52C9"
- "\u82B1"
- "\u30E4"
- "\u8EF8"
- "\u97FB"
- "\u66F2"
- "\u6574"
- "\u652F"
- "\u6271"
- "\u53E5"
- "\u6280"
- "\u5317"
- "\u30D8"
- "\u897F"
- "\u5247"
- "\u4FEE"
- "\u6388"
- "\u9031"
- "\u5BA4"
- "\u52D9"
- "\u9664"
- "\u533B"
- "\u6563"
- "\u56FA"
- "\u7AEF"
- "\u653E"
- "\u99AC"
- "\u7A4D"
- "\u8208"
- "\u592A"
- "\u5ACC"
- "\u9F62"
- "\u672B"
- "\u7D05"
- "\u6E90"
- "\u6E80"
- "\u5931"
- "\u5BDD"
- "\u6D88"
- "\u6E08"
- "\u4FBF"
- "\u983C"
- "\u4F01"
- "\u5B8C"
- "\u4F11"
- "\u9752"
- "\u7591"
- "\u8D70"
- "\u6975"
- "\u767B"
- "\u8AC7"
- "\u6839"
- "\u6025"
- "\u512A"
- "\u7D75"
- "\u623B"
- "\u5E2B"
- "\u5F59"
- "\u6DF7"
- "\u8DEF"
- "\u7E70"
- "\uFF2B"
- "\u8A3C"
- "\u713C"
- "\u6562"
- "\u5BB3"
- "\u96F6"
- "\u6253"
- "\u82E6"
- "\u7701"
- "\u7D19"
- "\u5C02"
- "\u8DDD"
- "\u9854"
- "\u8D8A"
- "\u4E89"
- "\u56F0"
- "\u5BC4"
- "\u5199"
- "\u4E92"
- "\u6DF1"
- "\u5A5A"
- "\u7DCF"
- "\u89A7"
- "\u80CC"
- "\u7BC9"
- "\u6E29"
- "\u8336"
- "\u62EC"
- "\u8CA0"
- "\u590F"
- "\u89E6"
- "\u7D14"
- "\u9045"
- "\u58EB"
- "\u96A3"
- "\u6050"
- "\u91C8"
- "\u967A"
- "\u5150"
- "\u5BBF"
- "\u6A21"
- "\u77F3"
- "\u983B"
- "\u5B09"
- "\u5EA7"
- "\u7642"
- "\u7E4B"
- "\uFF38"
- "\u5C06"
- "\u8FFD"
- "\u5EAD"
- "\u6238"
- "\u5371"
- "\u5BC6"
- "\u5DF1"
- "\u9014"
- "\u7BC4"
- "\u99C4"
- "\u7D39"
- "\u4EFB"
- "\u968F"
- "\u5357"
- "\uFF11"
- "\u5EB7"
- "\u9818"
- "\u5FD8"
- "\u3045"
- "\u59FF"
- "\u7F8E"
- "\u55B6"
- "\u6349"
- "\u65E2"
- "\u7167"
- "\uFF2A"
- "\u4EF2"
- "\u9152"
- "\u52E2"
- "\u9ED2"
- "\u5149"
- "\u6E21"
- "\u75DB"
- "\u62C5"
- "\u5F31"
- "\u307D"
- "\uFF36"
- "\u7D0D"
- "\u629C"
- "\u5E45"
- "\u6D17"
- "\u7A81"
- "\u671B"
- "\u5373"
- "\u9858"
- "\u7565"
- "\uFF12"
- "\u9811"
- "\u5FD7"
- "\u5B85"
- "\u7247"
- "\u656C"
- "\u6751"
- "\u60B2"
- "\u81A8"
- "\u89D2"
- "\u30E8"
- "\u4F9D"
- "\u8A73"
- "\u5F8B"
- "\u9B5A"
- "\u52B4"
- "\u5A66"
- "\u6163"
- "\u732B"
- "\u5019"
- "\u8001"
- "\u558B"
- "\u79F0"
- "\u796D"
- "\u7FA4"
- "\u7E2E"
- "\u6C38"
- "\u616E"
- "\u5EF6"
- "\u7A3F"
- "\u611B"
- "\u8089"
- "\u9589"
- "\u8CBB"
- "\u6295"
- "\u6D3E"
- "\u81F4"
- "\u7BA1"
- "\u7C73"
- "\u5E95"
- "\u7D99"
- "\u6C0F"
- "\u690D"
- "\u501F"
- "\u5727"
- "\u52E4"
- "\u6F22"
- "\u66AE"
- "\u5F27"
- "\u88C5"
- "\u57CE"
- "\u5287"
- "\u76DB"
- "\u63F4"
- "\u9244"
- "\u8C37"
- "\u5E72"
- "\u7E26"
- "\u8A31"
- "\u6016"
- "\u9A5A"
- "\u8A8C"
- "\uFF35"
- "\u8B77"
- "\u5B88"
- "\u8033"
- "\u6B32"
- "\u8239"
- "\uFF10"
- "\u5178"
- "\u67D3"
- "\u7D1A"
- "\u98FE"
- "\u5144"
- "\u71B1"
- "\u8F09"
- "\u88FD"
- "\u5BFA"
- "\u662D"
- "\u7FFB"
- "\u5426"
- "\u5584"
- "\u62BC"
- "\u53CA"
- "\u6A29"
- "\u559C"
- "\u670D"
- "\u8CB0"
- "\u8EFD"
- "\u677F"
- "\u61B6"
- "\u98FC"
- "\u5C3E"
- "\u5FA9"
- "\u5E78"
- "\u7389"
- "\u5354"
- "\u679A"
- "\u90CE"
- "\u8840"
- "\u524A"
- "\u5922"
- "\u63A1"
- "\u6674"
- "\u6B20"
- "\u602A"
- "\u65BD"
- "\u7DE8"
- "\u98EF"
- "\u7B56"
- "\u9000"
- "\uFF39"
- "\u8349"
- "\u61F8"
- "\u6458"
- "\u58CA"
- "\u4F38"
- "\u85AC"
- "\u9996"
- "\u5BFF"
- "\u53B3"
- "\u606F"
- "\u5C45"
- "\u643A"
- "\u9F3B"
- "\u9280"
- "\u4EA1"
- "\u6CCA"
- "\u8857"
- "\u9759"
- "\u9CE5"
- "\u677E"
- "\u5F92"
- "\u969C"
- "\u7B4B"
- "\u7559"
- "\u51B7"
- "\u5C24"
- "\u68EE"
- "\u5438"
- "\u5012"
- "\u68B0"
- "\u6D0B"
- "\u821E"
- "\u6A4B"
- "\u500D"
- "\u6255"
- "\u5352"
- "\u7E04"
- "\u6C5A"
- "\u53F8"
- "\u6625"
- "\u793C"
- "\u66DC"
- "\u6545"
- "\u526F"
- "\u5F01"
- "\u5439"
- "\u85E4"
- "\u8DE1"
- "\u962A"
- "\u4E86"
- "\u91E3"
- "\u9632"
- "\u7834"
- "\u6012"
- "\u662F"
- "\u30A5"
- "\u7AF6"
- "\u8179"
- "\u4E95"
- "\u4E08"
- "\u64AE"
- "\u72ED"
- "\u5BD2"
- "\u7B46"
- "\u5965"
- "\u8C4A"
- "\u732E"
- "\u5C31"
- "\u5A18"
- "\u79D2"
- "\u6C5F"
- "\u8E0F"
- "\u8A13"
- "\u7372"
- "\u96E8"
- "\u6BBA"
- "\u57CB"
- "\u64CD"
- "\u9AA8"
- "\u8D85"
- "\u6D5C"
- "\u8B66"
- "\u7DD1"
- "\u7D61"
- "\u8133"
- "\u7B11"
- "\u6D6E"
- "\u7D66"
- "\u7126"
- "\u8A70"
- "\u878D"
- "\u738B"
- "\u5C3A"
- "\u5E7C"
- "\u820C"
- "\u663C"
- "\u88CF"
- "\u6CE3"
- "\u67C4"
- "\u9396"
- "\u62E1"
- "\u8A3A"
- "\u7DE0"
- "\u5B98"
- "\u6697"
- "\u820E"
- "\u6298"
- "\u5264"
- "\u4E73"
- "\u6B6F"
- "\u7248"
- "\u5C04"
- "\u8108"
- "\u9707"
- "\u7802"
- "\u4F34"
- "\u72AF"
- "\u4F50"
- "\u5DDE"
- "\u8FB2"
- "\u8DA3"
- "\u990A"
- "\u675F"
- "\u6E2F"
- "\u8FEB"
- "\u5F3E"
- "\u798F"
- "\u51AC"
- "\u541B"
- "\u6B66"
- "\u77AC"
- "\u67A0"
- "\u6CA2"
- "\u661F"
- "\u5BCC"
- "\u6557"
- "\u5D0E"
- "\u6355"
- "\u8377"
- "\u5F1F"
- "\u95BE"
- "\u7E54"
- "\u7C89"
- "\u725B"
- "\u8DF5"
- "\u9999"
- "\u6797"
- "\u83DC"
- "\u62CD"
- "\u63CF"
- "\u888B"
- "\u6607"
- "\u91DD"
- "\u8FCE"
- "\u585A"
- "\u5A46"
- "\uFF49"
- "\u8ECD"
- "\uFF13"
- "\uFF37"
- "\u5BC2"
- "\u8F29"
- "\u3074"
- "\u5DFB"
- "\u4E01"
- "\u504F"
- "\u79CB"
- "\u5E9C"
- "\u6CC9"
- "\u81F3"
- "\u6368"
- "\u7956"
- "\u8584"
- "\u5B97"
- "\u5FB9"
- "\u93E1"
- "\u75C7"
- "\u6CB9"
- "\u8131"
- "\u9CF4"
- "\u7AE5"
- "\u6BDB"
- "\u9077"
- "\u84CB"
- "\u58C1"
- "\u5915"
- "\u5589"
- "\u907F"
- "\u984D"
- "\u6EA2"
- "\u96F0"
- "\u4EE4"
- "\u59C9"
- "\u63E1"
- "\u3077"
- "\u523B"
- "\u62E0"
- "\u8CA1"
- "\u8FF7"
- "\u9063"
- "\u82B8"
- "\u5E8F"
- "\u76E3"
- "\u8457"
- "\u5869"
- "\u5009"
- "\u7F6A"
- "\u6F5C"
- "\u7D5E"
- "\u764C"
- "\u5BAE"
- "\u5E2D"
- "\u8F2A"
- "\u594F"
- "\u846C"
- "\u6C60"
- "\u6CBF"
- "\u5FAE"
- "\u5305"
- "\u76CA"
- "\u76AE"
- "\u4FC3"
- "\u6297"
- "\u5FEB"
- "\u66AB"
- "\u52E7"
- "\u8CA9"
- "\u8C46"
- "\u5B63"
- "\u529F"
- "\u9A12"
- "\uFF54"
- "\u97D3"
- "\u6ED1"
- "\u75B2"
- "\u9003"
- "\u9061"
- "\u5E79"
- "\u60A9"
- "\u83D3"
- "\u672D"
- "\u6804"
- "\u9177"
- "\u8B1D"
- "\u6C96"
- "\u96EA"
- "\u5360"
- "\u60D1"
- "\u63FA"
- "\u866B"
- "\u62B1"
- "\uFF4B"
- "\u5CA1"
- "\u6E9C"
- "\u8535"
- "\u7763"
- "\u6838"
- "\u4E71"
- "\u4E45"
- "\u9EC4"
- "\u9670"
- "\u7720"
- "\u7B26"
- "\u6B8A"
- "\u628A"
- "\u6291"
- "\u5E0C"
- "\u63C3"
- "\u6483"
- "\u5EAB"
- "\u5409"
- "\u6E6F"
- "\u65CB"
- "\u640D"
- "\u52AA"
- "\u64E6"
- "\u9769"
- "\u6E0B"
- "\u773C"
- "\u592E"
- "\u8CDE"
- "\u5374"
- "\u5948"
- "\u539A"
- "\u59D4"
- "\u83EF"
- "\u96A0"
- "\uFF4E"
- "\u30CC"
- "\u9BAE"
- "\u515A"
- "\u5C65"
- "\u8A98"
- "\u6469"
- "\u6162"
- "\u5442"
- "\u7206"
- "\u7BB1"
- "\u6075"
- "\u9678"
- "\u7DCA"
- "\u7E3E"
- "\u5742"
- "\u7B52"
- "\u7532"
- "\u5348"
- "\u5230"
- "\u8CAC"
- "\u5C0A"
- "\u6CF3"
- "\u6279"
- "\u7518"
- "\u5B6B"
- "\u7159"
- "\u8A2A"
- "\u50B7"
- "\u6E05"
- "\u716E"
- "\u88C1"
- "\u9694"
- "\u8ED2"
- "\uFF31"
- "\u7FBD"
- "\u5D29"
- "\u7A74"
- "\u7CD6"
- "\u707D"
- "\u5275"
- "\u6F70"
- "\u6691"
- "\u87BA"
- "\u653B"
- "\u6577"
- "\u6575"
- "\u76E4"
- "\u9732"
- "\u7A93"
- "\u63B2"
- "\u81E8"
- "\u53E9"
- "\u5145"
- "\u4FFA"
- "\u8F38"
- "\u967D"
- "\u6B27"
- "\u6687"
- "\u6B6A"
- "\u6DFB"
- "\u60A3"
- "\u5FD9"
- "\u70AD"
- "\u829D"
- "\u8EDF"
- "\u88D5"
- "\u7E01"
- "\u6F2B"
- "\u7A1A"
- "\u7968"
- "\u8A69"
- "\u5CB8"
- "\u7687"
- "\uFF4A"
- "\u6627"
- "\u5100"
- "\u5857"
- "\u8E0A"
- "\u8AF8"
- "\u6D74"
- "\u904D"
- "\u66D6"
- "\u5BE7"
- "\u99B4"
- "\u5339"
- "\u03B1"
- "\u627F"
- "\u30BE"
- "\u6383"
- "\u5375"
- "\u5999"
- "\u3043"
- "\u66B4"
- "\u62B5"
- "\u604B"
- "\u8863"
- "\u6EB6"
- "\u7DAD"
- "\u514D"
- "\u6392"
- "\u685C"
- "\u7573"
- "\u7B87"
- "\u6398"
- "\u535A"
- "\u6FC3"
- "\u7FCC"
- "\u8056"
- "\u7DB2"
- "\u885B"
- "\u64EC"
- "\u5E8A"
- "\u9178"
- "\u6669"
- "\u4E7E"
- "\u90AA"
- "\u7551"
- "\u6EDE"
- "\u5802"
- "\u7E41"
- "\u4ECF"
- "\u5FB3"
- "\u7DE9"
- "\u6A39"
- "\u6551"
- "\u633F"
- "\u68D2"
- "\u906D"
- "\u676F"
- "\u6065"
- "\u6E56"
- "\u6E09"
- "\u81D3"
- "\u8CB4"
- "\u723A"
- "\u7981"
- "\u4F75"
- "\u5263"
- "\u786C"
- "\u58C7"
- "\u80A9"
- "\u6D78"
- "\u4F0A"
- "\u5B9D"
- "\u6094"
- "\u8E8D"
- "\u6DB2"
- "\u99C6"
- "\u6D25"
- "\u307A"
- "\u6D45"
- "\u8B72"
- "\u5CA9"
- "\u9B45"
- "\u587E"
- "\u03B8"
- "\u6696"
- "\u6CB3"
- "\u8A95"
- "\u7F36"
- "\u5507"
- "\u80A2"
- "\u6328"
- "\u62F6"
- "\u7A0E"
- "\u50AC"
- "\u8A34"
- "\uFF58"
- "\u968A"
- "\u659C"
- "\u770B"
- "\uFF50"
- "\u6D66"
- "\u8352"
- "\uFF41"
- "\u71C3"
- "\u52A3"
- "\u5BA3"
- "\u8FBF"
- "\u790E"
- "\u62FE"
- "\u5C4A"
- "\u6905"
- "\u5EC3"
- "\u6749"
- "\u9AEA"
- "\u77E2"
- "\u67D4"
- "\u55AB"
- "\u73CD"
- "\u57FC"
- "\u88C2"
- "\u63B4"
- "\u59BB"
- "\u8CA7"
- "\u934B"
- "\u59A5"
- "\u59B9"
- "\u5175"
- "\uFF14"
- "\u623F"
- "\u5951"
- "\u65E8"
- "\uFF44"
- "\u0394"
- "\u5DE1"
- "\u8A02"
- "\u5F90"
- "\u8CC0"
- "\u7BED"
- "\u9810"
- "\u84C4"
- "\u8846"
- "\u5DE8"
- "\u5506"
- "\u65E6"
- "\u5531"
- "\u9047"
- "\u6E67"
- "\u8010"
- "\u96C4"
- "\u6D99"
- "\u8CB8"
- "\u822A"
- "\u5104"
- "\u5618"
- "\u6C37"
- "\u78C1"
- "\u679D"
- "\u8CAB"
- "\u61D0"
- "\u52DF"
- "\u8155"
- "\u65E7"
- "\u7AF9"
- "\u99D0"
- "\u8A72"
- "\uFF52"
- "\u5893"
- "\u518A"
- "\u80F8"
- "\u758E"
- "\u773A"
- "\uFF45"
- "\u9855"
- "\u631F"
- "\u55A7"
- "\u520A"
- "\u68C4"
- "\u990C"
- "\u67F1"
- "\u5800"
- "\u8ACB"
- "\u79D8"
- "\u6717"
- "\u96F2"
- "\u8170"
- "\u7A32"
- "\u828B"
- "\u8C9D"
- "\u5C48"
- "\u91CC"
- "\u508D"
- "\u8102"
- "\u6FC1"
- "\u54B2"
- "\u6BD2"
- "\u6EC5"
- "\u5629"
- "\u6442"
- "\u6E7E"
- "\u83CC"
- "\u8150"
- "\u5211"
- "\u5F25"
- "\u5AC1"
- "\u61A7"
- "\u4E18"
- "\u5C90"
- "\u52B1"
- "\u8CA2"
- "\u6C41"
- "\u96C7"
- "\u5076"
- "\u9774"
- "\u72D9"
- "\u719F"
- "\u900F"
- "\uFF59"
- "\u8CFC"
- "\u5319"
- "\uFF46"
- "\uFF15"
- "\u92AD"
- "\u6D12"
- "\u8A17"
- "\u809D"
- "\u963F"
- "\u80C3"
- "\uFF53"
- "\u885D"
- "\u621A"
- "\uFF4D"
- "\u84B8"
- "\u4FF3"
- "\u8972"
- "\u5265"
- "\u5BE9"
- "\u6817"
- "\u8A87"
- "\u5237"
- "\u7CF8"
- "\u90F7"
- "\u5049"
- "\u6C57"
- "\u53CC"
- "\u98FD"
- "\u77DB"
- "\u984E"
- "\u552F"
- "\u6590"
- "\u7DB4"
- "\u5B64"
- "\u90F5"
- "\u76D7"
- "\u9E7F"
- "\u8CC3"
- "\u76FE"
- "\u682A"
- "\u9ED9"
- "\u7C8B"
- "\u63DA"
- "\u9808"
- "\u7092"
- "\u9285"
- "\u5E81"
- "\u9B54"
- "\u75E9"
- "\u9802"
- "\u76BF"
- "\u970A"
- "\u5E55"
- "\u570F"
- "\u574A"
- "\u72C2"
- "\u8912"
- "\u9451"
- "\u50B5"
- "\u77AD"
- "\u565B"
- "\u5E33"
- "\u5782"
- "\u8870"
- "\u4ED9"
- "\u9EA6"
- "\u8CA8"
- "\u7AAA"
- "\u6F6E"
- "\u6FEF"
- "\u5238"
- "\u7D1B"
- "\u7384"
- "\u7C4D"
- "\uFF43"
- "\u74F6"
- "\u5DE3"
- "\u5192"
- "\u6CBC"
- "\u99D2"
- "\u5C3D"
- "\u517C"
- "\u7C97"
- "\u63BB"
- "\u80BA"
- "\u9154"
- "\uFF4C"
- "\u702C"
- "\u505C"
- "\u6F20"
- "\u673A"
- "\u916C"
- "\u4FD7"
- "\u8986"
- "\u5C3B"
- "\u9375"
- "\u5805"
- "\u6F2C"
- "\u2212"
- "\u79C0"
- "\u6885"
- "\u9042"
- "\u57F9"
- "\u871C"
- "\uFF42"
- "\u30FB"
- "\u52C7"
- "\u8ECC"
- "\u7F85"
- "\uFF3A"
- "\u5BB4"
- "\u8C5A"
- "\u7A3C"
- "\u62AB"
- "\u8CAF"
- "\u9EBB"
- "\u6C4E"
- "\u51DD"
- "\u5FE0"
- "\uFF55"
- "\u5F80"
- "\u8AE6"
- "\u8B19"
- "\u6F0F"
- "\u5410"
- "\u3047"
- "\u7652"
- "\u9663"
- "\u6D6A"
- "\u52D8"
- "\u53D9"
- "\u5200"
- "\u67B6"
- "\u57F7"
- "\u5674"
- "\u5197"
- "\u4E4F"
- "\u837B"
- "\u81ED"
- "\u708A"
- "\u598A"
- "\u808C"
- "\u8CDB"
- "\u5C0B"
- "\u9175"
- "\u757F"
- "\u5270"
- "\u706F"
- "\u8C6A"
- "\u9685"
- "\u9905"
- "\u7949"
- "\u80AF"
- "\u62DB"
- "\u7A3D"
- "\u5F6B"
- "\u5F69"
- "\u03B2"
- "\u6B04"
- "\u718A"
- "\u68CB"
- "\u6CB8"
- "\u6C88"
- "\u8339"
- "\u7ABA"
- "\u5B9C"
- "\u8217"
- "\u7CA7"
- "\u683D"
- "\u80AA"
- "\u9665"
- "\u6CE1"
- "\u95D8"
- "\u8F3F"
- "\u5353"
- "\u7070"
- "\u8F9B"
- "\u6F01"
- "\u9F13"
- "\u585E"
- "\u8CD1"
- "\u76C6"
- "\u68FA"
- "\u6311"
- "\u54F2"
- "\u9867"
- "\u8B21"
- "\u8302"
- "\u90A3"
- "\u80DE"
- "\u4F3A"
- "\u5A92"
- "\u708E"
- "\u67D0"
- "\u564C"
- "\u5203"
- "\u6F5F"
- "\u7656"
- "\u4E80"
- "\u63EE"
- "\u511F"
- "\u4E39"
- "\u7DEF"
- "\u9DB4"
- "\u4E4B"
- "\u6BB4"
- "\u4EF0"
- "\u5949"
- "\u7E2B"
- "\u75F4"
- "\u8650"
- "\u61B2"
- "\u71E5"
- "\u6DC0"
- "\uFF57"
- "\u88F8"
- "\u82BD"
- "\u63A7"
- "\u95A3"
- "\u7587"
- "\u925B"
- "\u8178"
- "\u5642"
- "\u935B"
- "\u654F"
- "\u9162"
- "\u938C"
- "\u81E3"
- "\u8E74"
- "\u5A01"
- "\u6D44"
- "\u7965"
- "\u795D"
- "\u86C7"
- "\u811A"
- "\u4F0F"
- "\u6F54"
- "\u5510"
- "\u6955"
- "\u57A3"
- "\u932F"
- "\u514B"
- "\u614C"
- "\u6BBF"
- "\u819C"
- "\u61A9"
- "\u9065"
- "\u82DB"
- "\u9676"
- "\u8997"
- "\u78E8"
- "\u624D"
- "\u5E1D"
- "\u642C"
- "\u722A"
- "\u90CA"
- "\u80A5"
- "\u819D"
- "\u62D2"
- "\u868A"
- "\u5208"
- "\u5132"
- "\uFF48"
- "\u596E"
- "\u7761"
- "\u5BEE"
- "\uFF17"
- "\u4FB5"
- "\u9B31"
- "\u635C"
- "\u6DBC"
- "\u5A20"
- "\u7363"
- "\u7C92"
- "\u963B"
- "\u6CE5"
- "\u7ADC"
- "\u91A4"
- "\u92ED"
- "\u6606"
- "\u9234"
- "\u7DBF"
- "\u830E"
- "\u8107"
- "\u7948"
- "\u8A60"
- "\u6B53"
- "\u7F70"
- "\u68DA"
- "\u83CA"
- "\u6069"
- "\u7267"
- "\u540A"
- "\u8DF3"
- "\u6DE1"
- "\u7F72"
- "\u596A"
- "\u9038"
- "\u6170"
- "\u5EB6"
- "\u9262"
- "\u8B5C"
- "\u5ECA"
- "\u5606"
- "\u62ED"
- "\u8CED"
- "\u99C1"
- "\u7F8A"
- "\u5384"
- "\u7D10"
- "\u9673"
- "\u816B"
- "\u6841"
- "\u9298"
- "\u96CC"
- "\u636E"
- "\u62DD"
- "\u60E8"
- "\u96DB"
- "\u845B"
- "\u7FA8"
- "\u609F"
- "\u76DF"
- "\u7E4A"
- "\u9192"
- "\u65EC"
- "\u6DAF"
- "\u8CC4"
- "\u6E7F"
- "\u6F02"
- "\u7D2B"
- "\u30F4"
- "\u4E9C"
- "\u8AA0"
- "\u5854"
- "\u5E4C"
- "\u80C6"
- "\u64A5"
- "\u865A"
- "\u6F64"
- "\u9699"
- "\u5F84"
- "\u6C72"
- "\u8CE2"
- "\u5BF8"
- "\u8888"
- "\u88DF"
- "\u8266"
- "\uFF19"
- "\u62D8"
- "\uFF47"
- "\u5841"
- "\u5BDB"
- "\u51A0"
- "\u614E"
- "\u971E"
- "\u731B"
- "\u67CF"
- "\u733F"
- "\u9084"
- "\u50E7"
- "\u53EB"
- "\u53F1"
- "\u72E9"
- "\u63C9"
- "\u7D2F"
- "\u5982"
- "\u7897"
- "\u6BBB"
- "\u906E"
- "\u5FCD"
- "\u6EF4"
- "\u6B96"
- "\u8D08"
- "\u74A7"
- "\u6F38"
- "\u6589"
- "\u03BC"
- "\u9686"
- "\u6176"
- "\u72A0"
- "\u7272"
- "\u5146"
- "\u576A"
- "\u6284"
- "\u65D7"
- "\u50DA"
- "\u5C3F"
- "\u51CD"
- "\u902E"
- "\u7B39"
- "\u8F1D"
- "\u5C1A"
- "\u8015"
- "\u51CC"
- "\u632B"
- "\u4F10"
- "\u7BB8"
- "\u4E91"
- "\u5968"
- "\u819A"
- "\u9010"
- "\u03B3"
- "\u5F26"
- "\u9700"
- "\u5C01"
- "\u5E3D"
- "\u6F31"
- "\u9283"
- "\u507D"
- "\u5875"
- "\u7E1B"
- "\u58A8"
- "\u6020"
- "\u96F7"
- "\u5766"
- "\u68A8"
- "\u90ED"
- "\u7A4F"
- "\u67FF"
- "\u7AFF"
- "\u5E61"
- "\u5F81"
- "\u99B3"
- "\u9EBA"
- "\u03C4"
- "\u8154"
- "\u7C98"
- "\u7409"
- "\u731F"
- "\u4EC1"
- "\u8358"
- "\u6492"
- "\u7C3F"
- "\u90E1"
- "\u7B4C"
- "\u5D8B"
- "\u6FE1"
- "\u618E"
- "\u5446"
- "\u6F15"
- "\u5A29"
- "\u68DF"
- "\u6052"
- "\uFF18"
- "\u5553"
- "\u5B5D"
- "\u67F3"
- "\u64A4"
- "\u85CD"
- "\u95C7"
- "\u5B22"
- "\u67F4"
- "\u6734"
- "\u6D1E"
- "\u5CB3"
- "\u9B3C"
- "\u8DE8"
- "\u3049"
- "\u70C8"
- "\u559A"
- "\u6F84"
- "\u6FEB"
- "\u82A6"
- "\u62D3"
- "\u51FD"
- "\u6843"
- "\u76F2"
- "\u6CA1"
- "\u7A6B"
- "\u6212"
- "\u99FF"
- "\u8D05"
- "\u67AF"
- "\u6C70"
- "\u53F6"
- "\u90A6"
- "\u66C7"
- "\u9A30"
- "\u711A"
- "\u51F6"
- "\u5CF0"
- "\u69FD"
- "\u67DA"
- "\u5320"
- "\u9A19"
- "\u502B"
- "\u84EE"
- "\u634C"
- "\u61F2"
- "\u8B0E"
- "\u91B8"
- "\u56DA"
- "\u7344"
- "\u6EDD"
- "\u6795"
- "\u60DC"
- "\u7DB1"
- "\u8B33"
- "\u7089"
- "\u5DFE"
- "\u91DC"
- "\u9BAB"
- "\u6E58"
- "\u92F3"
- "\u5351"
- "\uFF51"
- "\u7DBB"
- "\u5EF7"
- "\u85A6"
- "\u667A"
- "\u6C99"
- "\u8CBF"
- "\u8098"
- "\uFF16"
- "\u5F0A"
- "\u66F0"
- "\u7881"
- "\u9DFA"
- "\u6676"
- "\u8D74"
- "\u8513"
- "\u75D2"
- "\u79E9"
- "\u5DE7"
- "\u9418"
- "\u7B1B"
- "\u638C"
- "\u53EC"
- "\u5347"
- "\u6249"
- "\u5A2F"
- "\u8A1F"
- "\u8247"
- "\u64B2"
- "\uFF56"
- "\u6182"
- "\u90B8"
- "\u5098"
- "\u7CDE"
- "\u03BB"
- "\u5C16"
- "\u723D"
- "\u7832"
- "\u55A9"
- "\u80CE"
- "\u84B2"
- "\u9DF9"
- "\u755C"
- "\u6897"
- "\uFF4F"
- "\u5023"
- "\u6247"
- "\u7DFB"
- "\u6756"
- "\u622F"
- "\u5D50"
- "\u6A3D"
- "\u6F06"
- "\u9CE9"
- "\u039B"
- "\u5FAA"
- "\u8896"
- "\u9784"
- "\u6851"
- "\u5D16"
- "\u59A8"
- "\u66A6"
- "\u59D3"
- "\u7A00"
- "\u3041"
- "\u920D"
- "\u9727"
- "\u9837"
- "\u8105"
- "\u7B20"
- "\u86CD"
- "\u8328"
- "\u69CD"
- "\u3062"
- "\u59EB"
- "\u6ABB"
- "\u8463"
- "\u6C7D"
- "\u541F"
- "\u807E"
- "\u73E0"
- "\u62B9"
- "\u9D28"
- "\u64AB"
- "\u8607"
- "\u7AC3"
- "\u864E"
- "\u78EF"
- "\u77E9"
- "\u7CCA"
- "\u55AA"
- "\u8A6E"
- "\u82D1"
- "\u98F4"
- "\u6089"
- "\u674F"
- "\u9B42"
- "\u914C"
- "\u9BC9"
- "\u8A50"
- "\u03A3"
- "\u7815"
- "\u55DC"
- "\u7FFC"
- "\u4F0E"
- "\u751A"
- "\u5F66"
- "\u961C"
- "\u8706"
- "\u6109"
- "\u80F4"
- "\u8776"
- "\u8B00"
- "\u9271"
- "\u75E2"
- "\u73ED"
- "\u9438"
- "\u92F8"
- "\u62D9"
- "\u6068"
- "\u4EAD"
- "\u4EAB"
- "\u75AB"
- "\u5F13"
- "\u74E6"
- "\u7D46"
- "\u814E"
- "\u62F3"
- "\u9A0E"
- "\u58B3"
- "\u83F1"
- "\u6813"
- "\u5256"
- "\u6D2A"
- "\u5484"
- "\u9591"
- "\u58EE"
- "\u9945"
- "\u65ED"
- "\u8987"
- "\u80A1"
- "\u86D9"
- "\u724C"
- "\u965B"
- "\u714E"
- "\u63AC"
- "\u9AED"
- "\u9019"
- "\u5E7B"
- "\u54B3"
- "\u6E26"
- "\u55C5"
- "\u7A42"
- "\u7434"
- "\u5FCC"
- "\u70CF"
- "\u5448"
- "\u91D8"
- "\u611A"
- "\u6C3E"
- "\u8AFE"
- "\u6E9D"
- "\u7336"
- "\u7AAF"
- "\u8ACF"
- "\u8CC2"
- "\u57C3"
- "\u51F8"
- "\u7D0B"
- "\u6ADB"
- "\u525B"
- "\u98E2"
- "\u4FCA"
- "\u54C0"
- "\u5BB0"
- "\u93AE"
- "\u7435"
- "\u7436"
- "\u96C5"
- "\u8494"
- "\u85AA"
- "\u8A93"
- "\u59EA"
- "\u62D7"
- "\u8778"
- "\u7169"
- "\u7B51"
- "\u690E"
- "\u4FB6"
- "\u553E"
- "\u7BAA"
- "\u5075"
- "\u8861"
- "\u03C3"
- "\u88FE"
- "\u95B2"
- "\u805A"
- "\u4E3C"
- "\u633D"
- "\u7E4D"
- "\u82D7"
- "\u9E93"
- "\u03C6"
- "\u03B4"
- "\u4E32"
- "\u51E1"
- "\u5F18"
- "\u85FB"
- "\u61C7"
- "\u817F"
- "\u7A9F"
- "\u6803"
- "\u6652"
- "\u5E84"
- "\u7891"
- "\u7B4F"
- "\u7B25"
- "\u5E06"
- "\u96B7"
- "\u8FB0"
- "\u75BE"
- "\u8FE6"
- "\u8A6B"
- "\u5617"
- "\u582A"
- "\u6842"
- "\u5B9B"
- "\u58F7"
- "\u8AED"
- "\u97AD"
- "\u9310"
- "\u6DF5"
- "\u79E4"
- "\u7525"
- "\u4F8D"
- "\u66FD"
- "\u6572"
- "\u63AA"
- "\u6168"
- "\u83E9"
- "\u5CE0"
- "\u901D"
- "\u5F70"
- "\u67F5"
- "\u82AF"
- "\u7C50"
- "\u57A2"
- "\u03BE"
- "\u77EF"
- "\u8C8C"
- "\u8F44"
- "\u8A89"
- "\u9813"
- "\u7D79"
- "\u9E78"
- "\u5E7D"
- "\u6881"
- "\u642D"
- "\u54BD"
- "\u82B3"
- "\u7729"
- "\u0393"
- "\u61A4"
- "\u7985"
- "\u6063"
- "\u5840"
- "\u7149"
- "\u75FA"
- "\uFF06"
- "\u7A40"
- "\u545F"
- "\u918D"
- "\u9190"
- "\u7901"
- "\u51F9"
- "\u86EE"
- "\u5974"
- "\u64AD"
- "\u7E79"
- "\u8499"
- "\u8A63"
- "\u4E5F"
- "\u5420"
- "\u4E59"
- "\u8E8A"
- "\u8E87"
- "\u9D2C"
- "\u7A92"
- "\u59E5"
- "\u9326"
- "\u694A"
- "\u8017"
- "\u6F09"
- "\u60E7"
- "\u4FE3"
- "\u6876"
- "\u5CFB"
- "\u905C"
- "\u65FA"
- "\u75D5"
- "\u03A6"
- "\u6234"
- "\u658E"
- "\u8CD3"
- "\u7BC7"
- "\u8429"
- "\u85E9"
- "\u7950"
- "\u8B83"
- "\u83AB"
- "\u9C39"
- "\u85A9"
- "\u5378"
- "\u4E9B"
- "\u75B9"
- "\u8E44"
- "\u4E56"
- "\uFF5A"
- "\u92FC"
- "\u6A3A"
- "\u5B8F"
- "\u7BE4"
- "\u8258"
- "\u81B3"
- "\u7A83"
- "\u7E82"
- "\u5598"
- "\u786B"
- "\u99D5"
- "\u7261"
- "\u732A"
- "\u62D0"
- "\u60DA"
- "\u60A0"
- "\u7CE7"
- "\u95A5"
- "\u03C0"
- "\u853D"
- "\u6850"
- "\u981A"
- "\u9214"
- "\u697C"
- "\u8C9E"
- "\u602F"
- "\u817A"
- "\u8305"
- "\u6CF0"
- "\u9913"
- "\u5C51"
- "\u9BDB"
- "\u929B"
- "\u9AB8"
- "\u9C57"
- "\u5824"
- "\u9675"
- "\u6DD8"
- "\u64C1"
- "\u81FC"
- "\u6D32"
- "\u8FBB"
- "\u8A23"
- "\u5C4F"
- "\u9BE8"
- "\u895F"
- "\u5CE1"
- "\u660C"
- "\u982C"
- "\u5806"
- "\u865C"
- "\u840E"
- "\u9EB9"
- "\u7CE0"
- "\u68B1"
- "\u8AFA"
- "\u5403"
- "\u66A2"
- "\u5B54"
- "\u5EB8"
- "\u5DF3"
- "\u589C"
- "\u85AE"
- "\u6101"
- "\u664B"
- "\u8236"
- "\u8FC5"
- "\u6B3A"
- "\u9640"
- "\u7709"
- "\u6CC4"
- "\u59FB"
- "\u9688"
- "\u58CC"
- "\u69D9"
- "\u5E87"
- "\u52D2"
- "\u6E07"
- "\u91E7"
- "\u4E43"
- "\u82D4"
- "\u9306"
- "\u58D5"
- "\u78D0"
- "\u6962"
- "\u65A7"
- "\u5E63"
- "\u03B7"
- "\u7E55"
- "\u83C5"
- "\u7109"
- "\u5112"
- "\u5D07"
- "\u8276"
- "\u5449"
- "\u7984"
- "\u54C9"
- "\u68AF"
- "\u5937"
- "\u546A"
- "\u56C3"
- "\u84BC"
- "\u9A28"
- "\u9D3B"
- "\u862D"
- "\u7CA5"
- "\u7D3A"
- "\u7D17"
- "\u7164"
- "\u03C9"
- "\u52FE"
- "\u97A0"
- "\u4F3D"
- "\u7AAE"
- "\u6E15"
- "\u0392"
- "\u8D66"
- "\u6597"
- "\u66F9"
- "\u8CE0"
- "\u5CAC"
- "\u847A"
- "\u7D33"
- "\u5B8D"
- "\u6191"
- "\u6357"
- "\u7C9B"
- "\u8CCA"
- "\u9F8D"
- "\u81C6"
- "\u6C8C"
- "\u52C5"
- "\u8096"
- "\u559D"
- "\u8CAA"
- "\u82AD"
- "\u8549"
- "\u919C"
- "\u64B9"
- "\u5740"
- "\u7BE0"
- "\u7D2C"
- "\u75B1"
- "\u52F2"
- "\u86FE"
- "\u88B4"
- "\u8749"
- "\u685F"
- "\u4FF5"
- "\u818F"
- "\u5DF7"
- "\u5072"
- "\u6148"
- "\u754F"
- "\u96BB"
- "\u606D"
- "\u64B0"
- "\u9D0E"
- "\u52AB"
- "\u63C6"
- "\u914E"
- "\u8106"
- "\u6241"
- "\u9761"
- "\u8511"
- "\u95CA"
- "\u96BC"
- "\u6CCC"
- "\u5996"
- "\u65A1"
- "\u52C3"
- "\u637B"
- "\u6E13"
- "\u937E"
- "\u5954"
- "\u6155"
- "\u5984"
- "\u6A0B"
- "\u936C"
- "\u502D"
- "\u8679"
- "\u03BD"
- "\u60A6"
- "\u8151"
- "\u62EE"
- "\u51E0"
- "\u80E1"
- "\u8FC2"
- "\u8EAF"
- "\u50ED"
- "\u6ECB"
- "\u7B8B"
- "\u75F0"
- "\u65AC"
- "\u85AB"
- "\u673D"
- "\u82A5"
- "\u9756"
- "\u907C"
- "\u6591"
- "\u7953"
- "\u5B95"
- "\u976D"
- "\u72D7"
- "\u81BF"
- "\u59AC"
- "\u5A7F"
- "\u7554"
- "\u7AEA"
- "\u9D5C"
- "\u8CE6"
- "\u7E1E"
- "\u6731"
- "\u7C95"
- "\u69FB"
- "\u6D69"
- "\u511A"
- "\u8CDC"
- "\u8B39"
- "\u68B5"
- "\u5A9B"
- "\u7947"
- "\u5516"
- "\u03C8"
- "\u03C1"
- "\u5A9A"
- "\u540E"
- "\u6FB1"
- "\u7DBE"
- "\u6372"
- "\u67E9"
- "\u6DF3"
- "\u74DC"
- "\u5631"
- "\u51B4"
- "\u6115"
- "\u9211"
- "\u51B6"
- "\u67A2"
- "\u03A9"
- "\u77B0"
- "\u6775"
- "\u5EB5"
- "\u4F2F"
- "\u840C"
- "\u5609"
- "\u4FC4"
- "\u7D06"
- "\u81A0"
- "\u7252"
- "\u8EB0"
- "\u543E"
- "\u50FB"
- "\u704C"
- "\u646F"
- "\u5091"
- "\u929A"
- "\u8B90"
- "\u8910"
- "\u8FB1"
- "\u7345"
- "\u7B94"
- "\u73A9"
- "\u4F43"
- "\u583A"
- "\u5504"
- "\u515C"
- "\u62CC"
- "\u5751"
- "\u75D8"
- "\u69CC"
- "\u77B3"
- "\u79BF"
- "\u66D9"
- "\u5DF2"
- "\u7FC1"
- "\u5C3C"
- "\u60BC"
- "\u7F77"
- "\u699C"
- "\u5451"
- "\u79E6"
- "\u533F"
- "\u03BA"
- "\u7259"
- "\u4F46"
- "\u572D"
- "\u548E"
- "\u745E"
- "\u7A1C"
- "\u785D"
- "\u6BC5"
- "\u7015"
- "\u8702"
- "\u978D"
- "\u6A2B"
- "\u7566"
- "\u660F"
- "\u755D"
- "\u4FAE"
- "\u548B"
- "\u6367"
- "\u7F9E"
- "\u803D"
- "\u60B8"
- "\u51E7"
- "\u4EAE"
- "\u9AC4"
- "\u54FA"
- "\u4FEF"
- "\u567A"
- "\u8058"
- "\u8654"
- "\u5B8B"
- "\u93A7"
- "\u968B"
- "\u51B3"
- "\u59D1"
- "\u7078"
- "\u927E"
- "\u8F5F"
- "\u60F0"
- "\u03C7"
- "\u643E"
- "\u6854"
- "\u7F6B"
- "\u8E4A"
- "\u68B6"
- "\u6893"
- "\u7F75"
- "\u65A5"
- "\u6276"
- "\u6147"
- "\u61C3"
- "\u9949"
- "\u6E25"
- "\u6AD3"
- "\u80E4"
- "\u56A2"
- "\u9CF3"
- "\u6A84"
- "\u8C79"
- "\u50B2"
- "\u50D1"
- "\u7586"
- "\u6134"
- "\u53A8"
- "\u6FB9"
- "\u9320"
- "\u64E2"
- "\u6EBA"
- "\u7624"
- "\u73CA"
- "\u5BC5"
- "\u6977"
- "\u9583"
- "\u9CF6"
- "\u7119"
- "\u6912"
- "\u9B4F"
- "\u9798"
- "\u68A2"
- "\u6900"
- "\u8ACC"
- "\u696B"
- "\u5F14"
- "\u65D2"
- "\u5957"
- "\u9F5F"
- "\u9F6C"
- "\u7D18"
- "\u810A"
- "\u536F"
- "\u727D"
- "\u6BD8"
- "\u6714"
- "\u514E"
- "\u721B"
- "\u6D9C"
- "\u5851"
- "\u5F04"
- "\u676D"
- "\u63A0"
- "\u80B4"
- "\u626E"
- "\u51F1"
- "\u798D"
- "\u8036"
- "\u808B"
- "\u7235"
- "\u61AB"
- "\u57D3"
- "\u5983"
- "\u9910"
- "\u7C7E"
- "\u7262"
- "\u6816"
- "\u9017"
- "\u7058"
- "\u5E5F"
- "\u68F2"
- "\u5687"
- "\u7827"
- "\u6E1A"
- "\u7C9F"
- "\u7A7F"
- "\u7F60"
- "\u68F9"
- "\u8594"
- "\u8587"
- "\u526A"
- "\u7B48"
- "\u936E"
- "\u892A"
- "\u7AA9"
- "\u58F1"
- "\u30F2"
- "\u7460"
- "\u7483"
- "\u61BE"
- "\u5E16"
- "\u6960"
- "\u03B5"
- "\u5480"
- "\u56BC"
- "\u56A5"
- "\u6D29"
- "\u6A58"
- "\u6867"
- "\u6A9C"
- "\u63F6"
- "\u63C4"
- "\u88E1"
- "\u6A80"
- "\u900D"
- "\u9081"
- "\u6028"
- "\u73B2"
- "\u90C1"
- "\u5815"
- "\u8AB9"
- "\u8B17"
- "\u8956"
- "\u51F0"
- "\u9B41"
- "\u5B75"
- "\u7766"
- "\u71FB"
- "\u5243"
- "\u53A9"
- "\u71D7"
- "\u84D1"
- "\u5EFB"
- "\u75D4"
- "\u837C"
- "\u6190"
- "\u6070"
- "\u8F9F"
- "\u5F98"
- "\u5F8A"
- "\u4FA0"
- "\u5830"
- "\u971C"
- "\u809B"
- "\u76E7"
- "\u5835"
- "\u72DB"
- "\u9D8F"
- "\u9119"
- "\u4F73"
- "\u916A"
- "\u8AE7"
- "\u6973"
- "\u7826"
- "\u5AC9"
- "\u5DEB"
- "\u53E1"
- "\u9716"
- "\u6E23"
- "\u5544"
- "\u798E"
- "\u6CAB"
- "\u821F"
- "\u6C5D"
- "\u5302"
- "\u99F1"
- "\u6C08"
- "\u308E"
- "\u714C"
- "\u7DAC"
- "\u5F1B"
- "\u586B"
- "\u84C1"
- "\u5039"
- "\u7CFE"
- "\u51A5"
- "\u674E"
- "\u966A"
- "\u8877"
- "\u59E6"
- "\u5962"
- "\u75BC"
- "\u8A54"
- "\u8599"
- "\u8B5A"
- "\u5CEF"
- "\u684E"
- "\u688F"
- "\u9B92"
- "\u8A1B"
- "\u55B0"
- "\u7960"
- "\u67A1"
- "\u6681"
- "\u4E5E"
- "\u91C7"
- "\u9739"
- "\u9742"
- "\u687F"
- "\u929C"
- "\u4F51"
- "\u79BE"
- "\u5944"
- "\u6930"
- "\u87F9"
- "\u8061"
- "\u98AF"
- "\u30C2"
- "\u8E81"
- "\u8E42"
- "\u8E99"
- "\u8695"
- "\u693F"
- "\u62F7"
- "\u9257"
- "\u8882"
- "\u78CB"
- "\u7422"
- "\u6B3D"
- "\u60B6"
- "\u53C9"
- "\u7E37"
- "\u8A36"
- "\u50C5"
- "\u5C6F"
- "\u5EEC"
- "\u5C41"
- "\u99A8"
- "\u6E20"
- "\u8568"
- "\u699B"
- "\u675C"
- "\u7791"
- "\u6A8E"
- "\u8ECB"
- "\u8F62"
- "\u8700"
- "\u8235"
- "\u82B9"
- "\u6B3E"
- "\u639F"
- "\u8E2A"
- "\u745A"
- "\u71E6"
- "\u7D21"
- "\u584A"
- "\u8171"
- "\u6753"
- "\u65A4"
- "\u786F"
- "\u55AC"
- "\u8B04"
- "\u79DF"
- "\u8180"
- "\u80F1"
- "\u6EC4"
- "\u9C10"
- "\u8475"
- "\u8471"
- "\u8461"
- "\u5A49"
- "\u88D4"
- "\u9F0E"
- "\u9187"
- "\u67EF"
- "\u991E"
- "\u96C1"
- "\u8AA6"
- "\u8A62"
- "\u633A"
- "\u7AFA"
- "\u8A82"
- "\u5191"
- "\u8718"
- "\u86DB"
- "\u70B8"
- "\u932B"
- "\u58C5"
- "\u8087"
- "\u54AC"
- "\u9B8E"
- "\u67D1"
- "\u7D9C"
- "\u5BE1"
- "\u7977"
- "\u522E"
- "\u8CCE"
- "\u9B18"
- "\u884D"
- "\u5FD6"
- "\u685D"
- "\u0398"
- "\u039A"
- "\u03A8"
- "\u53E2"
- "\u4FCE"
- "\u7396"
- "\u78A7"
- "\u8766"
- "\u8521"
- "\u649A"
- "\u7A14"
- "\u752B"
- "\u6D35"
- "\u7893"
- "\u9ECE"
- "\u5AE1"
- "\u8755"
- "\u725F"
- "\u6B89"
- "\u6C83"
- "\u7B50"
- "\u619A"
- "\u6E24"
- "\u9B4D"
- "\u9B4E"
- "\u71ED"
- "\u7940"
- "\u6D1B"
- "\u88F3"
- "\u4E11"
- "\u9846"
- "\u9952"
- "\u5EC9"
- "\u689F"
- "\u848B"
- "\u6DD1"
- "\u8737"
- "\u9644"
- "\u695A"
- "\u9F20"
- "\u5154"
- "\u61AC"
- "\u5F57"
- "\u66FC"
- "\u5D11"
- "\u57DC"
- "\u5F77"
- "\u5F7F"
- "\u5DF4"
- "\u831C"
- "\u6D9B"
- "\u57E0"
- "\u945A"
- "\u92D2"
- "\u5C09"
- "\u53AD"
- "\u7B75"
- "\u7AE3"
- "\u7E8F"
- "\u6194"
- "\u60B4"
- "\u8E5F"
- "\u675E"
- "\u7825"
- "\u8F14"
- "\u9C52"
- "\u4FAF"
- "\u7D62"
- "\u5475"
- "\u698E"
- "\u53EA"
- "\u71D5"
- "\u5C60"
- "\u5614"
- "\u74E2"
- "\u9291"
- "\u880D"
- "\u932C"
- "\u608C"
- "\u8A1D"
- "\u7DB8"
- "\u530D"
- "\u5310"
- "\u637A"
- "\u6A59"
- "\u5BB5"
- "\u9D60"
- "\u57F4"
- "\u7690"
- "\u9021"
- "\u4FF8"
- "\u7A63"
- "\u54A4"
- "\u8309"
- "\u8389"
- "\u6643"
- "\u6EF8"
- "\u5289"
- "\u5026"
- "\u8944"
- "\u7B4D"
- "\u5239"
- "\u83BD"
- "\u9041"
- "\u66F5"
- "\u79BD"
- "\u7B67"
- "\u7E0A"
- "\u7FD4"
- "\u5BF5"
- "\u834F"
- "\u758B"
- "\u84EC"
- "\u83B1"
- "\u8EAC"
- "\u696E"
- "\u76C8"
- "\u5C13"
- "\u72FC"
- "\u85C9"
- "\u965F"
- "\u620E"
- "\u4E8E"
- "\u6F58"
- "\u8012"
- "\u5F82"
- "\u5FA0"
- "\u99AE"
- "\u5F6D"
- "\u5E47"
- "\u9087"
- "\u6CD3"
- "\u80B1"
- "\u65BC"
- "\u6602"
- "\u8E64"
- "\u7463"
- "\u9A65"
- "\u4EA8"
- "\u8AEE"
- "\u77EE"
- "\u8569"
- "\u6566"
- "\u30EE"
- "\u6208"
- "\u8229"
- "\u9B6F"
- "\u65E0"
- "\u6159"
- "\u6127"
- "\u8340"
- "\u6309"
- "\u914B"
- "\u59F6"
- "\u723E"
- "\u8602"
- "\u986B"
- "\u593E"
- "\u59DA"
- "\u701D"
- "\u6FD8"
- "\u964B"
- "\u777E"
- "\u5B30"
- "\u5DBA"
- "\u821B"
- "\u7B65"
- "\u95A4"
- "\u68D8"
- "\u9812"
- "\u59BE"
- "\u8B2C"
- "\u4F0D"
- "\u537F"
- "\u8FEA"
- "\u5686"
- "\u60F9"
- "\u80DA"
- "\u6C6A"
- "\u543B"
- "\u9B51"
- "\u8F3B"
- "\u59C6"
- "\u84FC"
- "\u6AC2"
- "\u5315"
- "\u4F70"
- "\u7246"
- "\u5CD9"
- "\u725D"
- "\u9DF2"
- "\u7DCB"
- "\u7BAD"
- "\u82EB"
- "\u5366"
- "\u5B5F"
- "\u5323"
- "\u4ED4"
- "\u5D19"
- "\u6787"
- "\u6777"
- "\u81C0"
- "\u681E"
- "\u9E1E"
- "\u61FA"
- "\u55DA"
- "\u6DB8"
- "\u30C5"
- "\u8D16"
- "\u5E9A"
- "\u93D1"
- "\u9149"
- "\u670B"
- "\u70F9"
- "\u53C8"
- "\u7337"
- "\u7C00"
- "\u5B2C"
- "\u88B7"
- "\u6BB7"
- "\u51DB"
- "\u4EC0"
- "\u71FF"
- "\u5556"
- "\u7BC6"
- "\u7DD8"
- "\u5036"
- "\u6AC3"
- "\u8A03"
- "\u540F"
- "\u5CB1"
- "\u8A25"
- "\u958F"
- "\u5DBD"
- "\u722C"
- "\u618A"
- "\u7511"
- "\u6144"
- "\u5E25"
- "\u7704"
- "\u5A11"
- "\u50E5"
- "\u5016"
- "\u800C"
- "\u8F4D"
- "\u5583"
- "\u81BE"
- "\u7099"
- "\u85AF"
- "\u97EE"
- "\u4E99"
- "\u8B14"
- "\u86CE"
- "\u7425"
- "\u73C0"
- "\u698A"
- "\u7C3E"
- "\u8D6D"
- "\u8823"
- "\u8299"
- "\u8B01"
- "\u9022"
- "\u8466"
- "\u6670"
- "\u5398"
- "\u707C"
- "\u903C"
- "\u9328"
- "\u700B"
- "\u5FF8"
- "\u6029"
- "\u7165"
- "\u7B0F"
- "\u5FFD"
- "\u7708"
- "\u7DEC"
- "\u5C4D"
- "\u75BD"
- "\u6E5B"
- "\u788D"
- "\u8AE4"
- <sos/eos>
init: xavier_uniform
input_size: null
ctc_conf:
dropout_rate: 0.0
ctc_type: builtin
reduce: true
model_conf:
ctc_weight: 0.3
lsm_weight: 0.1
length_normalized_loss: false
use_preprocessor: true
token_type: char
bpemodel: null
non_linguistic_symbols: null
cleaner: null
g2p: null
frontend: default
frontend_conf:
fs: 16k
specaug: specaug
specaug_conf:
apply_time_warp: true
time_warp_window: 5
time_warp_mode: bicubic
apply_freq_mask: true
freq_mask_width_range:
- 0
- 30
num_freq_mask: 2
apply_time_mask: true
time_mask_width_range:
- 0
- 40
num_time_mask: 2
normalize: global_mvn
normalize_conf:
stats_file: exp/asr_stats_raw_sp/train/feats_stats.npz
encoder: conformer
encoder_conf:
output_size: 512
attention_heads: 8
linear_units: 2048
num_blocks: 12
dropout_rate: 0.1
positional_dropout_rate: 0.1
attention_dropout_rate: 0.1
input_layer: conv2d6
normalize_before: true
macaron_style: false
pos_enc_layer_type: rel_pos
selfattention_layer_type: rel_selfattn
activation_type: swish
use_cnn_module: true
cnn_module_kernel: 31
decoder: transformer
decoder_conf:
attention_heads: 8
linear_units: 2048
num_blocks: 6
dropout_rate: 0.1
positional_dropout_rate: 0.1
self_attention_dropout_rate: 0.1
src_attention_dropout_rate: 0.1
required:
- output_dir
- token_list
distributed: true
```
</details>
### Citing ESPnet
```BibTex
@inproceedings{watanabe2018espnet,
author={Shinji Watanabe and Takaaki Hori and Shigeki Karita and Tomoki Hayashi and Jiro Nishitoba and Yuya Unno and Nelson Yalta and Jahn Heymann and Matthew Wiesner and Nanxin Chen and Adithya Renduchintala and Tsubasa Ochiai},
title={{ESPnet}: End-to-End Speech Processing Toolkit},
year={2018},
booktitle={Proceedings of Interspeech},
pages={2207--2211},
doi={10.21437/Interspeech.2018-1456},
url={http://dx.doi.org/10.21437/Interspeech.2018-1456}
}
```
or arXiv:
```bibtex
@misc{watanabe2018espnet,
title={ESPnet: End-to-End Speech Processing Toolkit},
author={Shinji Watanabe and Takaaki Hori and Shigeki Karita and Tomoki Hayashi and Jiro Nishitoba and Yuya Unno and Nelson Yalta and Jahn Heymann and Matthew Wiesner and Nanxin Chen and Adithya Renduchintala and Tsubasa Ochiai},
year={2018},
eprint={1804.00015},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
```
|
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: "Kรฒ sรญ แบนฬrรญ tรญ รณ fi แบนsแบนฬ rinlแบนฬ ."
---
# 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, highยญ 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 personโs name right after another personโs name
I-PER |Personโs 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 = "Kรฒ sรญ แบนฬrรญ tรญ รณ fi แบนsแบนฬ rinlแบนฬ ."
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, highยญ 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 personโs name right after another personโs name
I-PER |Personโs 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 BรEY Cรฉy 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, highยญ 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 personโs name right after another personโs name
I-PER |Personโs 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 BรEY Cรฉy Koronaa !"
ner_results = nlp(example)
print(ner_results)
```
|
mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-yoruba | mbeukman | 2021-11-25T09:05:08Z | 8 | 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: "Kรฒ sรญ แบนฬrรญ tรญ รณ fi แบนsแบนฬ rinlแบนฬ ."
---
# xlm-roberta-base-finetuned-swahili-finetuned-ner-yoruba
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 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, highยญ 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 personโs name right after another personโs name
I-PER |Personโs 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-yoruba](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-yoruba) (This model) | [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-yoruba-finetuned-ner-yoruba](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-yoruba-finetuned-ner-yoruba) | [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-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-swahili-finetuned-ner-yoruba'
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
example = "Kรฒ sรญ แบนฬrรญ tรญ รณ fi แบนsแบนฬ rinlแบนฬ ."
ner_results = nlp(example)
print(ner_results)
```
|
mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-swahili | mbeukman | 2021-11-25T09:05:03Z | 33 | 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-swahili-finetuned-ner-swahili
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 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, highยญ 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 personโs name right after another personโs name
I-PER |Personโs 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-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-swahili) (This model) | [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-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-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-swahili-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-naija | mbeukman | 2021-11-25T09:05:00Z | 9 | 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-swahili-finetuned-ner-naija
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 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, highยญ 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 personโs name right after another personโs name
I-PER |Personโs 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-naija](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-naija) (This model) | [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-naija-finetuned-ner-naija](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-naija-finetuned-ner-naija) | [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-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-swahili-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-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: "๏ปฟJii 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, highยญ 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 personโs name right after another personโs name
I-PER |Personโs 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 = "๏ปฟJii 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-igbo | mbeukman | 2021-11-25T09:04:50Z | 5 | 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 แปda jแปฅแปฅ otแปฅ nkeji banyere oke ogbugbu na - eme n'ala Naijiria agwแปฅla Ekweremmadแปฅ"
---
# xlm-roberta-base-finetuned-swahili-finetuned-ner-igbo
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 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, highยญ 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 personโs name right after another personโs name
I-PER |Personโs 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-igbo](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-swahili-finetuned-ner-igbo) (This model) | [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-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-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-swahili-finetuned-ner-igbo'
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
example = "Ike แปda jแปฅแปฅ otแปฅ nkeji banyere oke ogbugbu na - eme n'ala Naijiria agwแปฅla Ekweremmadแปฅ"
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, highยญ 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 personโs name right after another personโs name
I-PER |Personโs 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-yoruba | mbeukman | 2021-11-25T09:04:45Z | 4 | 2 | 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: "Kรฒ sรญ แบนฬrรญ tรญ รณ fi แบนsแบนฬ rinlแบนฬ ."
---
# xlm-roberta-base-finetuned-ner-yoruba
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 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, highยญ 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 personโs name right after another personโs name
I-PER |Personโs 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-yoruba](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-ner-yoruba) (This model) | [base](https://huggingface.co/xlm-roberta-base) | yor | 78.22 | 77.21 | 79.26 | 77.00 | 80.00 | 71.00 | 82.00 |
| [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-yoruba-finetuned-ner-yoruba](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-yoruba-finetuned-ner-yoruba) | [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 |
## 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-yoruba'
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
example = "Kรฒ sรญ แบนฬrรญ tรญ รณ fi แบนsแบนฬ rinlแบนฬ ."
ner_results = nlp(example)
print(ner_results)
```
|
mbeukman/xlm-roberta-base-finetuned-ner-wolof | mbeukman | 2021-11-25T09:04:43Z | 7 | 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 BรEY Cรฉy Koronaa !"
---
# xlm-roberta-base-finetuned-ner-wolof
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 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, highยญ 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 personโs name right after another personโs name
I-PER |Personโs 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-wolof](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-ner-wolof) (This model) | [base](https://huggingface.co/xlm-roberta-base) | wol | 66.12 | 69.46 | 63.09 | 30.00 | 84.00 | 54.00 | 59.00 |
| [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-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 |
## 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-wolof'
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
example = "SAFIYETU BรEY Cรฉy Koronaa !"
ner_results = nlp(example)
print(ner_results)
```
|
mbeukman/xlm-roberta-base-finetuned-ner-swahili | mbeukman | 2021-11-25T09:04:40Z | 7 | 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-ner-swahili
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 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, highยญ 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 personโs name right after another personโs name
I-PER |Personโs 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-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-ner-swahili) (This model) | [base](https://huggingface.co/xlm-roberta-base) | swa | 88.71 | 86.84 | 90.67 | 83.00 | 91.00 | 79.00 | 95.00 |
| [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-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 |
## 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-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-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: "๏ปฟJii 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, highยญ 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 personโs name right after another personโs name
I-PER |Personโs 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 = "๏ปฟJii 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, highยญ 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 personโs name right after another personโs name
I-PER |Personโs 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 โ 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, highยญ 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 personโs name right after another personโs name
I-PER |Personโs 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 โ 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 แปda jแปฅแปฅ otแปฅ nkeji banyere oke ogbugbu na - eme n'ala Naijiria agwแปฅla Ekweremmadแปฅ"
---
# 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, highยญ 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 personโs name right after another personโs name
I-PER |Personโs 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 แปda jแปฅแปฅ otแปฅ nkeji banyere oke ogbugbu na - eme n'ala Naijiria agwแปฅla Ekweremmadแปฅ"
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, highยญ 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 personโs name right after another personโs name
I-PER |Personโs 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-kinyarwanda-finetuned-ner-swahili | mbeukman | 2021-11-25T09:04:07Z | 4 | 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-kinyarwanda-finetuned-ner-swahili
This is a token classification (specifically NER) model that fine-tuned [xlm-roberta-base-finetuned-kinyarwanda](https://huggingface.co/Davlan/xlm-roberta-base-finetuned-kinyarwanda) 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, highยญ 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 personโs name right after another personโs name
I-PER |Personโs 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-kinyarwanda-finetuned-ner-swahili](https://huggingface.co/mbeukman/xlm-roberta-base-finetuned-kinyarwanda-finetuned-ner-swahili) (This model) | [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-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-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-kinyarwanda-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-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, highยญ 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 personโs name right after another personโs name
I-PER |Personโs 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 แปda jแปฅแปฅ otแปฅ nkeji banyere oke ogbugbu na - eme n'ala Naijiria agwแปฅla Ekweremmadแปฅ"
---
# 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, highยญ 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 personโs name right after another personโs name
I-PER |Personโs 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 แปda jแปฅแปฅ otแปฅ nkeji banyere oke ogbugbu na - eme n'ala Naijiria agwแปฅla Ekweremmadแปฅ"
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, highยญ 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 personโs name right after another personโs name
I-PER |Personโs 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)
```
|
jpabbuehl/distilbert-base-uncased-finetuned-cola | jpabbuehl | 2021-11-25T08:49:51Z | 6 | 0 | transformers | [
"transformers",
"pytorch",
"tensorboard",
"distilbert",
"text-classification",
"generated_from_trainer",
"dataset:glue",
"license:apache-2.0",
"model-index",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text-classification | 2022-03-02T23:29:05Z | ---
license: apache-2.0
tags:
- generated_from_trainer
datasets:
- glue
metrics:
- matthews_correlation
model-index:
- name: distilbert-base-uncased-finetuned-cola
results:
- task:
name: Text Classification
type: text-classification
dataset:
name: glue
type: glue
args: cola
metrics:
- name: Matthews Correlation
type: matthews_correlation
value: 0.5229586822934302
---
<!-- 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-cola
This model is a fine-tuned version of [distilbert-base-uncased](https://huggingface.co/distilbert-base-uncased) on the glue dataset.
It achieves the following results on the evaluation set:
- Loss: 0.7588
- Matthews Correlation: 0.5230
## 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: 5
### Training results
| Training Loss | Epoch | Step | Validation Loss | Matthews Correlation |
|:-------------:|:-----:|:----:|:---------------:|:--------------------:|
| 0.5261 | 1.0 | 535 | 0.5125 | 0.4124 |
| 0.3502 | 2.0 | 1070 | 0.5439 | 0.5076 |
| 0.2378 | 3.0 | 1605 | 0.6629 | 0.4946 |
| 0.1809 | 4.0 | 2140 | 0.7588 | 0.5230 |
| 0.1309 | 5.0 | 2675 | 0.8901 | 0.5056 |
### Framework versions
- Transformers 4.12.5
- Pytorch 1.10.0+cu111
- Datasets 1.15.1
- Tokenizers 0.10.3
|
DiegoAlysson/opus-mt-en-ro-finetuned-en-to-ro | DiegoAlysson | 2021-11-25T03:08:55Z | 8 | 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 | 2022-03-02T23:29:04Z | ---
license: apache-2.0
tags:
- generated_from_trainer
datasets:
- wmt16
metrics:
- bleu
model-index:
- name: opus-mt-en-ro-finetuned-en-to-ro
results:
- task:
name: Sequence-to-sequence Language Modeling
type: text2text-generation
dataset:
name: wmt16
type: wmt16
args: ro-en
metrics:
- name: Bleu
type: bleu
value: 27.9273
---
<!-- 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-en-ro-finetuned-en-to-ro
This model is a fine-tuned version of [Helsinki-NLP/opus-mt-en-ro](https://huggingface.co/Helsinki-NLP/opus-mt-en-ro) on the wmt16 dataset.
It achieves the following results on the evaluation set:
- Loss: 1.2915
- Bleu: 27.9273
- Gen Len: 34.0935
## 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 |
|:-------------:|:-----:|:-----:|:---------------:|:-------:|:-------:|
| 0.7448 | 1.0 | 38145 | 1.2915 | 27.9273 | 34.0935 |
### Framework versions
- Transformers 4.12.5
- Pytorch 1.10.0+cu111
- Datasets 1.15.1
- Tokenizers 0.10.3
|
arnolfokam/bert-base-uncased-pcm | arnolfokam | 2021-11-24T21:14:03Z | 7 | 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
**bert-base-uncased-pcm** is a model based on the fine-tuned 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
-|-|-|-
**bert-base-uncased-pcm**| 88.61 | 84.17 | 86.33
# Usage
```python
from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
tokenizer = AutoTokenizer.from_pretrained("arnolfokam/bert-base-uncased-pcm")
model = AutoModelForTokenClassification.from_pretrained("arnolfokam/bert-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)
``` |
huggingtweets/emirtarik | huggingtweets | 2021-11-24T20:31:24Z | 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/emirtarik/1637785880110/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/1435194184294707207/s3hAS9Pv_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">๐ค AI BOT ๐ค</div>
<div style="text-align: center; font-size: 16px; font-weight: 800">Emir</div>
<div style="text-align: center; font-size: 14px;">@emirtarik</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 Emir.
| Data | Emir |
| --- | --- |
| Tweets downloaded | 1917 |
| Retweets | 421 |
| Short tweets | 368 |
| Tweets kept | 1128 |
[Explore the data](https://wandb.ai/wandb/huggingtweets/runs/2bk4sb83/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 @emirtarik's tweets.
Hyperparameters and metrics are recorded in the [W&B training run](https://wandb.ai/wandb/huggingtweets/runs/3abibhtt) for full transparency and reproducibility.
At the end of training, [the final model](https://wandb.ai/wandb/huggingtweets/runs/3abibhtt/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/emirtarik')
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)
|
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 ๐ค"
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-large-msmarco-10k | castorini | 2021-11-24T19:15:14Z | 149 | 1 | 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-large reranker fine-tuned on the MS MARCO passage dataset for 10k steps (or 1 epoch).
This model usually has a better zero-shot performance than `monot5-large-msmarco`, i.e., it performs better on datasets different from MS MARCO.
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/) |
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-yelp_polarity | AdapterHub | 2021-11-24T16:33:21Z | 1 | 0 | adapter-transformers | [
"adapter-transformers",
"text-classification",
"roberta",
"en",
"dataset:yelp_polarity",
"arxiv:2104.08247",
"region:us"
] | text-classification | 2022-03-02T23:29:04Z | ---
tags:
- text-classification
- roberta
- adapter-transformers
datasets:
- yelp_polarity
language:
- en
---
# Adapter `AdapterHub/roberta-base-pf-yelp_polarity` for roberta-base
An [adapter](https://adapterhub.ml) for the `roberta-base` model that was trained on the [yelp_polarity](https://huggingface.co/datasets/yelp_polarity/) 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-yelp_polarity", 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-wnut_17 | AdapterHub | 2021-11-24T16:33:15Z | 5 | 0 | adapter-transformers | [
"adapter-transformers",
"token-classification",
"roberta",
"en",
"dataset:wnut_17",
"arxiv:2104.08247",
"region:us"
] | token-classification | 2022-03-02T23:29:04Z | ---
tags:
- token-classification
- roberta
- adapter-transformers
datasets:
- wnut_17
language:
- en
---
# Adapter `AdapterHub/roberta-base-pf-wnut_17` for roberta-base
An [adapter](https://adapterhub.ml) for the `roberta-base` model that was trained on the [wnut_17](https://huggingface.co/datasets/wnut_17/) 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-wnut_17", 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-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-snli | AdapterHub | 2021-11-24T16:31:56Z | 2 | 0 | adapter-transformers | [
"adapter-transformers",
"text-classification",
"roberta",
"en",
"dataset:snli",
"arxiv:2104.08247",
"region:us"
] | text-classification | 2022-03-02T23:29:04Z | ---
tags:
- text-classification
- roberta
- adapter-transformers
datasets:
- snli
language:
- en
---
# Adapter `AdapterHub/roberta-base-pf-snli` for roberta-base
An [adapter](https://adapterhub.ml) for the `roberta-base` model that was trained on the [snli](https://huggingface.co/datasets/snli/) 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-snli", 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-rotten_tomatoes | AdapterHub | 2021-11-24T16:31:26Z | 2 | 0 | adapter-transformers | [
"adapter-transformers",
"text-classification",
"roberta",
"adapterhub:sentiment/rotten_tomatoes",
"en",
"dataset:rotten_tomatoes",
"arxiv:2104.08247",
"region:us"
] | text-classification | 2022-03-02T23:29:04Z | ---
tags:
- text-classification
- roberta
- adapterhub:sentiment/rotten_tomatoes
- adapter-transformers
datasets:
- rotten_tomatoes
language:
- en
---
# Adapter `AdapterHub/roberta-base-pf-rotten_tomatoes` for roberta-base
An [adapter](https://adapterhub.ml) for the `roberta-base` model that was trained on the [sentiment/rotten_tomatoes](https://adapterhub.ml/explore/sentiment/rotten_tomatoes/) 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-rotten_tomatoes", 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-race | AdapterHub | 2021-11-24T16:31:15Z | 6 | 0 | adapter-transformers | [
"adapter-transformers",
"adapterhub:rc/race",
"roberta",
"en",
"dataset:race",
"arxiv:2104.08247",
"region:us"
] | null | 2022-03-02T23:29:04Z | ---
tags:
- adapterhub:rc/race
- roberta
- adapter-transformers
datasets:
- race
language:
- en
---
# Adapter `AdapterHub/roberta-base-pf-race` for roberta-base
An [adapter](https://adapterhub.ml) for the `roberta-base` model that was trained on the [rc/race](https://adapterhub.ml/explore/rc/race/) 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-race", 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 Rรผcklรฉ 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-mrpc | AdapterHub | 2021-11-24T16:30:24Z | 2 | 0 | adapter-transformers | [
"adapter-transformers",
"text-classification",
"roberta",
"adapterhub:sts/mrpc",
"en",
"arxiv:2104.08247",
"region:us"
] | text-classification | 2022-03-02T23:29:04Z | ---
tags:
- text-classification
- roberta
- adapterhub:sts/mrpc
- adapter-transformers
language:
- en
---
# Adapter `AdapterHub/roberta-base-pf-mrpc` for roberta-base
An [adapter](https://adapterhub.ml) for the `roberta-base` model that was trained on the [sts/mrpc](https://adapterhub.ml/explore/sts/mrpc/) 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-mrpc", 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-imdb | AdapterHub | 2021-11-24T16:30:07Z | 49 | 0 | adapter-transformers | [
"adapter-transformers",
"text-classification",
"roberta",
"adapterhub:sentiment/imdb",
"en",
"dataset:imdb",
"arxiv:2104.08247",
"region:us"
] | text-classification | 2022-03-02T23:29:04Z | ---
tags:
- text-classification
- roberta
- adapterhub:sentiment/imdb
- adapter-transformers
datasets:
- imdb
language:
- en
---
# Adapter `AdapterHub/roberta-base-pf-imdb` for roberta-base
An [adapter](https://adapterhub.ml) for the `roberta-base` model that was trained on the [sentiment/imdb](https://adapterhub.ml/explore/sentiment/imdb/) 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-imdb", 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-hellaswag | AdapterHub | 2021-11-24T16:29:56Z | 3 | 0 | adapter-transformers | [
"adapter-transformers",
"roberta",
"adapterhub:comsense/hellaswag",
"en",
"dataset:hellaswag",
"arxiv:2104.08247",
"region:us"
] | null | 2022-03-02T23:29:04Z | ---
tags:
- roberta
- adapterhub:comsense/hellaswag
- adapter-transformers
datasets:
- hellaswag
language:
- en
---
# Adapter `AdapterHub/roberta-base-pf-hellaswag` for roberta-base
An [adapter](https://adapterhub.ml) for the `roberta-base` model that was trained on the [comsense/hellaswag](https://adapterhub.ml/explore/comsense/hellaswag/) 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-hellaswag", 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 Rรผcklรฉ 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-cosmos_qa | AdapterHub | 2021-11-24T16:29:17Z | 2 | 0 | adapter-transformers | [
"adapter-transformers",
"roberta",
"adapterhub:comsense/cosmosqa",
"en",
"dataset:cosmos_qa",
"arxiv:2104.08247",
"region:us"
] | null | 2022-03-02T23:29:04Z | ---
tags:
- roberta
- adapterhub:comsense/cosmosqa
- adapter-transformers
datasets:
- cosmos_qa
language:
- en
---
# Adapter `AdapterHub/roberta-base-pf-cosmos_qa` for roberta-base
An [adapter](https://adapterhub.ml) for the `roberta-base` model that was trained on the [comsense/cosmosqa](https://adapterhub.ml/explore/comsense/cosmosqa/) 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-cosmos_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 Rรผcklรฉ 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-conll2000 | AdapterHub | 2021-11-24T16:28:49Z | 5 | 0 | adapter-transformers | [
"adapter-transformers",
"token-classification",
"roberta",
"adapterhub:chunk/conll2000",
"en",
"dataset:conll2000",
"arxiv:2104.08247",
"region:us"
] | token-classification | 2022-03-02T23:29:04Z | ---
tags:
- token-classification
- roberta
- adapterhub:chunk/conll2000
- adapter-transformers
datasets:
- conll2000
language:
- en
---
# Adapter `AdapterHub/roberta-base-pf-conll2000` for roberta-base
An [adapter](https://adapterhub.ml) for the `roberta-base` model that was trained on the [chunk/conll2000](https://adapterhub.ml/explore/chunk/conll2000/) 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-conll2000", 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-cola | AdapterHub | 2021-11-24T16:27:49Z | 11 | 1 | adapter-transformers | [
"adapter-transformers",
"text-classification",
"roberta",
"adapterhub:lingaccept/cola",
"en",
"arxiv:2104.08247",
"region:us"
] | text-classification | 2022-03-02T23:29:04Z | ---
tags:
- text-classification
- roberta
- adapterhub:lingaccept/cola
- adapter-transformers
language:
- en
---
# Adapter `AdapterHub/roberta-base-pf-cola` for roberta-base
An [adapter](https://adapterhub.ml) for the `roberta-base` model that was trained on the [lingaccept/cola](https://adapterhub.ml/explore/lingaccept/cola/) 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-cola", 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-art | AdapterHub | 2021-11-24T16:27:34Z | 1 | 0 | adapter-transformers | [
"adapter-transformers",
"roberta",
"en",
"dataset:art",
"arxiv:2104.08247",
"region:us"
] | null | 2022-03-02T23:29:04Z | ---
tags:
- roberta
- adapter-transformers
datasets:
- art
language:
- en
---
# Adapter `AdapterHub/roberta-base-pf-art` for roberta-base
An [adapter](https://adapterhub.ml) for the `roberta-base` model that was trained on the [art](https://huggingface.co/datasets/art/) 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-art", 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 Rรผcklรฉ 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/bert-base-uncased-pf-yelp_polarity | AdapterHub | 2021-11-24T16:27:20Z | 4 | 0 | adapter-transformers | [
"adapter-transformers",
"text-classification",
"bert",
"en",
"dataset:yelp_polarity",
"arxiv:2104.08247",
"region:us"
] | text-classification | 2022-03-02T23:29:04Z | ---
tags:
- text-classification
- bert
- adapter-transformers
datasets:
- yelp_polarity
language:
- en
---
# Adapter `AdapterHub/bert-base-uncased-pf-yelp_polarity` for bert-base-uncased
An [adapter](https://adapterhub.ml) for the `bert-base-uncased` model that was trained on the [yelp_polarity](https://huggingface.co/datasets/yelp_polarity/) 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("bert-base-uncased")
adapter_name = model.load_adapter("AdapterHub/bert-base-uncased-pf-yelp_polarity", 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/bert-base-uncased-pf-wnut_17 | AdapterHub | 2021-11-24T16:27:13Z | 3 | 0 | adapter-transformers | [
"adapter-transformers",
"token-classification",
"bert",
"en",
"dataset:wnut_17",
"arxiv:2104.08247",
"region:us"
] | token-classification | 2022-03-02T23:29:04Z | ---
tags:
- token-classification
- bert
- adapter-transformers
datasets:
- wnut_17
language:
- en
---
# Adapter `AdapterHub/bert-base-uncased-pf-wnut_17` for bert-base-uncased
An [adapter](https://adapterhub.ml) for the `bert-base-uncased` model that was trained on the [wnut_17](https://huggingface.co/datasets/wnut_17/) 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("bert-base-uncased")
adapter_name = model.load_adapter("AdapterHub/bert-base-uncased-pf-wnut_17", 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/bert-base-uncased-pf-winogrande | AdapterHub | 2021-11-24T16:27:05Z | 0 | 0 | adapter-transformers | [
"adapter-transformers",
"bert",
"adapterhub:comsense/winogrande",
"en",
"dataset:winogrande",
"arxiv:2104.08247",
"region:us"
] | null | 2022-03-02T23:29:04Z | ---
tags:
- bert
- adapterhub:comsense/winogrande
- adapter-transformers
datasets:
- winogrande
language:
- en
---
# Adapter `AdapterHub/bert-base-uncased-pf-winogrande` for bert-base-uncased
An [adapter](https://adapterhub.ml) for the `bert-base-uncased` model that was trained on the [comsense/winogrande](https://adapterhub.ml/explore/comsense/winogrande/) 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("bert-base-uncased")
adapter_name = model.load_adapter("AdapterHub/bert-base-uncased-pf-winogrande", 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 Rรผcklรฉ 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/bert-base-uncased-pf-wic | AdapterHub | 2021-11-24T16:26:54Z | 1 | 0 | adapter-transformers | [
"adapter-transformers",
"text-classification",
"bert",
"adapterhub:wordsence/wic",
"en",
"arxiv:2104.08247",
"region:us"
] | text-classification | 2022-03-02T23:29:04Z | ---
tags:
- text-classification
- bert
- adapterhub:wordsence/wic
- adapter-transformers
language:
- en
---
# Adapter `AdapterHub/bert-base-uncased-pf-wic` for bert-base-uncased
An [adapter](https://adapterhub.ml) for the `bert-base-uncased` model that was trained on the [wordsence/wic](https://adapterhub.ml/explore/wordsence/wic/) 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("bert-base-uncased")
adapter_name = model.load_adapter("AdapterHub/bert-base-uncased-pf-wic", 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/bert-base-uncased-pf-ud_pos | AdapterHub | 2021-11-24T16:26:47Z | 11 | 0 | adapter-transformers | [
"adapter-transformers",
"token-classification",
"bert",
"adapterhub:pos/ud_ewt",
"en",
"dataset:universal_dependencies",
"arxiv:2104.08247",
"region:us"
] | token-classification | 2022-03-02T23:29:04Z | ---
tags:
- token-classification
- bert
- adapterhub:pos/ud_ewt
- adapter-transformers
datasets:
- universal_dependencies
language:
- en
---
# Adapter `AdapterHub/bert-base-uncased-pf-ud_pos` for bert-base-uncased
An [adapter](https://adapterhub.ml) for the `bert-base-uncased` 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("bert-base-uncased")
adapter_name = model.load_adapter("AdapterHub/bert-base-uncased-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/bert-base-uncased-pf-social_i_qa | AdapterHub | 2021-11-24T16:26:06Z | 4 | 0 | adapter-transformers | [
"adapter-transformers",
"bert",
"en",
"dataset:social_i_qa",
"arxiv:2104.08247",
"region:us"
] | null | 2022-03-02T23:29:04Z | ---
tags:
- bert
- adapter-transformers
datasets:
- social_i_qa
language:
- en
---
# Adapter `AdapterHub/bert-base-uncased-pf-social_i_qa` for bert-base-uncased
An [adapter](https://adapterhub.ml) for the `bert-base-uncased` 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("bert-base-uncased")
adapter_name = model.load_adapter("AdapterHub/bert-base-uncased-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 Rรผcklรฉ 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/bert-base-uncased-pf-scicite | AdapterHub | 2021-11-24T16:25:39Z | 2 | 0 | adapter-transformers | [
"adapter-transformers",
"text-classification",
"bert",
"en",
"dataset:scicite",
"arxiv:2104.08247",
"region:us"
] | text-classification | 2022-03-02T23:29:04Z | ---
tags:
- text-classification
- bert
- adapter-transformers
datasets:
- scicite
language:
- en
---
# Adapter `AdapterHub/bert-base-uncased-pf-scicite` for bert-base-uncased
An [adapter](https://adapterhub.ml) for the `bert-base-uncased` model that was trained on the [scicite](https://huggingface.co/datasets/scicite/) 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("bert-base-uncased")
adapter_name = model.load_adapter("AdapterHub/bert-base-uncased-pf-scicite", 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/bert-base-uncased-pf-race | AdapterHub | 2021-11-24T16:25:17Z | 4 | 0 | adapter-transformers | [
"adapter-transformers",
"adapterhub:rc/race",
"bert",
"en",
"dataset:race",
"arxiv:2104.08247",
"region:us"
] | null | 2022-03-02T23:29:04Z | ---
tags:
- adapterhub:rc/race
- bert
- adapter-transformers
datasets:
- race
language:
- en
---
# Adapter `AdapterHub/bert-base-uncased-pf-race` for bert-base-uncased
An [adapter](https://adapterhub.ml) for the `bert-base-uncased` model that was trained on the [rc/race](https://adapterhub.ml/explore/rc/race/) 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("bert-base-uncased")
adapter_name = model.load_adapter("AdapterHub/bert-base-uncased-pf-race", 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 Rรผcklรฉ 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/bert-base-uncased-pf-quail | AdapterHub | 2021-11-24T16:24:59Z | 1 | 0 | adapter-transformers | [
"adapter-transformers",
"bert",
"en",
"dataset:quail",
"arxiv:2104.08247",
"region:us"
] | null | 2022-03-02T23:29:04Z | ---
tags:
- bert
- adapter-transformers
datasets:
- quail
language:
- en
---
# Adapter `AdapterHub/bert-base-uncased-pf-quail` for bert-base-uncased
An [adapter](https://adapterhub.ml) for the `bert-base-uncased` model that was trained on the [quail](https://huggingface.co/datasets/quail/) 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("bert-base-uncased")
adapter_name = model.load_adapter("AdapterHub/bert-base-uncased-pf-quail", 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 Rรผcklรฉ 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/bert-base-uncased-pf-mnli | AdapterHub | 2021-11-24T16:24:19Z | 4 | 0 | adapter-transformers | [
"adapter-transformers",
"text-classification",
"bert",
"adapterhub:nli/multinli",
"en",
"dataset:multi_nli",
"arxiv:2104.08247",
"region:us"
] | text-classification | 2022-03-02T23:29:04Z | ---
tags:
- text-classification
- bert
- adapterhub:nli/multinli
- adapter-transformers
datasets:
- multi_nli
language:
- en
---
# Adapter `AdapterHub/bert-base-uncased-pf-mnli` for bert-base-uncased
An [adapter](https://adapterhub.ml) for the `bert-base-uncased` model that was trained on the [nli/multinli](https://adapterhub.ml/explore/nli/multinli/) 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("bert-base-uncased")
adapter_name = model.load_adapter("AdapterHub/bert-base-uncased-pf-mnli", 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/bert-base-uncased-pf-imdb | AdapterHub | 2021-11-24T16:24:07Z | 6 | 0 | adapter-transformers | [
"adapter-transformers",
"text-classification",
"bert",
"adapterhub:sentiment/imdb",
"en",
"dataset:imdb",
"arxiv:2104.08247",
"region:us"
] | text-classification | 2022-03-02T23:29:04Z | ---
tags:
- text-classification
- bert
- adapterhub:sentiment/imdb
- adapter-transformers
datasets:
- imdb
language:
- en
---
# Adapter `AdapterHub/bert-base-uncased-pf-imdb` for bert-base-uncased
An [adapter](https://adapterhub.ml) for the `bert-base-uncased` model that was trained on the [sentiment/imdb](https://adapterhub.ml/explore/sentiment/imdb/) 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("bert-base-uncased")
adapter_name = model.load_adapter("AdapterHub/bert-base-uncased-pf-imdb", 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/bert-base-uncased-pf-hellaswag | AdapterHub | 2021-11-24T16:23:47Z | 0 | 0 | adapter-transformers | [
"adapter-transformers",
"bert",
"adapterhub:comsense/hellaswag",
"en",
"dataset:hellaswag",
"arxiv:2104.08247",
"region:us"
] | null | 2022-03-02T23:29:04Z | ---
tags:
- bert
- adapterhub:comsense/hellaswag
- adapter-transformers
datasets:
- hellaswag
language:
- en
---
# Adapter `AdapterHub/bert-base-uncased-pf-hellaswag` for bert-base-uncased
An [adapter](https://adapterhub.ml) for the `bert-base-uncased` model that was trained on the [comsense/hellaswag](https://adapterhub.ml/explore/comsense/hellaswag/) 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("bert-base-uncased")
adapter_name = model.load_adapter("AdapterHub/bert-base-uncased-pf-hellaswag", 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 Rรผcklรฉ 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/bert-base-uncased-pf-emo | AdapterHub | 2021-11-24T16:23:01Z | 4 | 0 | adapter-transformers | [
"adapter-transformers",
"text-classification",
"bert",
"en",
"dataset:emo",
"arxiv:2104.08247",
"region:us"
] | text-classification | 2022-03-02T23:29:04Z | ---
tags:
- text-classification
- bert
- adapter-transformers
datasets:
- emo
language:
- en
---
# Adapter `AdapterHub/bert-base-uncased-pf-emo` for bert-base-uncased
An [adapter](https://adapterhub.ml) for the `bert-base-uncased` model that was trained on the [emo](https://huggingface.co/datasets/emo/) 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("bert-base-uncased")
adapter_name = model.load_adapter("AdapterHub/bert-base-uncased-pf-emo", 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/bert-base-uncased-pf-conll2003_pos | AdapterHub | 2021-11-24T16:22:26Z | 12 | 0 | adapter-transformers | [
"adapter-transformers",
"token-classification",
"bert",
"adapterhub:pos/conll2003",
"en",
"dataset:conll2003",
"arxiv:2104.08247",
"region:us"
] | token-classification | 2022-03-02T23:29:04Z | ---
tags:
- token-classification
- bert
- adapterhub:pos/conll2003
- adapter-transformers
datasets:
- conll2003
language:
- en
---
# Adapter `AdapterHub/bert-base-uncased-pf-conll2003_pos` for bert-base-uncased
An [adapter](https://adapterhub.ml) for the `bert-base-uncased` model that was trained on the [pos/conll2003](https://adapterhub.ml/explore/pos/conll2003/) 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("bert-base-uncased")
adapter_name = model.load_adapter("AdapterHub/bert-base-uncased-pf-conll2003_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/bert-base-uncased-pf-conll2003 | AdapterHub | 2021-11-24T16:22:18Z | 5 | 1 | adapter-transformers | [
"adapter-transformers",
"token-classification",
"bert",
"adapterhub:ner/conll2003",
"en",
"dataset:conll2003",
"arxiv:2104.08247",
"region:us"
] | token-classification | 2022-03-02T23:29:04Z | ---
tags:
- token-classification
- bert
- adapterhub:ner/conll2003
- adapter-transformers
datasets:
- conll2003
language:
- en
---
# Adapter `AdapterHub/bert-base-uncased-pf-conll2003` for bert-base-uncased
An [adapter](https://adapterhub.ml) for the `bert-base-uncased` model that was trained on the [ner/conll2003](https://adapterhub.ml/explore/ner/conll2003/) 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("bert-base-uncased")
adapter_name = model.load_adapter("AdapterHub/bert-base-uncased-pf-conll2003", 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/bert-base-uncased-pf-conll2000 | AdapterHub | 2021-11-24T16:22:12Z | 2 | 0 | adapter-transformers | [
"adapter-transformers",
"token-classification",
"bert",
"adapterhub:chunk/conll2000",
"en",
"dataset:conll2000",
"arxiv:2104.08247",
"region:us"
] | token-classification | 2022-03-02T23:29:04Z | ---
tags:
- token-classification
- bert
- adapterhub:chunk/conll2000
- adapter-transformers
datasets:
- conll2000
language:
- en
---
# Adapter `AdapterHub/bert-base-uncased-pf-conll2000` for bert-base-uncased
An [adapter](https://adapterhub.ml) for the `bert-base-uncased` model that was trained on the [chunk/conll2000](https://adapterhub.ml/explore/chunk/conll2000/) 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("bert-base-uncased")
adapter_name = model.load_adapter("AdapterHub/bert-base-uncased-pf-conll2000", 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",
}
``` |
Lowin/chinese-bigbird-mini-1024 | Lowin | 2021-11-24T16:05:17Z | 127 | 1 | transformers | [
"transformers",
"pytorch",
"big_bird",
"fill-mask",
"zh",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | fill-mask | 2022-03-02T23:29:04Z | ---
language:
- zh
license:
- apache-2.0
---
```python
import jieba_fast
from transformers import BertTokenizer
from transformers import BigBirdModel
class JiebaTokenizer(BertTokenizer):
def __init__(
self, pre_tokenizer=lambda x: jieba_fast.cut(x, HMM=False), *args, **kwargs
):
super().__init__(*args, **kwargs)
self.pre_tokenizer = pre_tokenizer
def _tokenize(self, text, *arg, **kwargs):
split_tokens = []
for text in self.pre_tokenizer(text):
if text in self.vocab:
split_tokens.append(text)
else:
split_tokens.extend(super()._tokenize(text))
return split_tokens
model = BigBirdModel.from_pretrained('Lowin/chinese-bigbird-mini-1024')
tokenizer = JiebaTokenizer.from_pretrained('Lowin/chinese-bigbird-mini-1024')
```
https://github.com/LowinLi/chinese-bigbird |
Lowin/chinese-bigbird-tiny-1024 | Lowin | 2021-11-24T16:03:15Z | 52 | 2 | transformers | [
"transformers",
"pytorch",
"big_bird",
"feature-extraction",
"zh",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | feature-extraction | 2022-03-02T23:29:04Z | ---
language:
- zh
license:
- apache-2.0
---
```python
import jieba_fast
from transformers import BertTokenizer
from transformers import BigBirdModel
class JiebaTokenizer(BertTokenizer):
def __init__(
self, pre_tokenizer=lambda x: jieba_fast.cut(x, HMM=False), *args, **kwargs
):
super().__init__(*args, **kwargs)
self.pre_tokenizer = pre_tokenizer
def _tokenize(self, text, *arg, **kwargs):
split_tokens = []
for text in self.pre_tokenizer(text):
if text in self.vocab:
split_tokens.append(text)
else:
split_tokens.extend(super()._tokenize(text))
return split_tokens
model = BigBirdModel.from_pretrained('Lowin/chinese-bigbird-tiny-1024')
tokenizer = JiebaTokenizer.from_pretrained('Lowin/chinese-bigbird-tiny-1024')
```
https://github.com/LowinLi/chinese-bigbird |
arnolfokam/bert-base-uncased-swa | arnolfokam | 2021-11-24T11:55:34Z | 10 | 0 | transformers | [
"transformers",
"pytorch",
"bert",
"token-classification",
"NER",
"swa",
"dataset:masakhaner",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:05Z | ---
language:
- swa
tags:
- NER
datasets:
- masakhaner
metrics:
- f1
- precision
- recall
license: apache-2.0
widget:
- text: "Wizara ya afya ya Tanzania imeripoti Jumatatu kuwa, watu takriban 14 zaidi wamepata maambukizi ya Covid-19."
---
# Model description
**bert-base-uncased-swa** is a model based on the fine-tuned 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 Swahili corpus **(swa)** 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 **(swa)** 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
-|-|-|-
**bert-base-uncased-swa**| 83.38 | 89.32 | 86.26
# Usage
```python
from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
tokenizer = AutoTokenizer.from_pretrained("arnolfokam/bert-base-uncased-swa")
model = AutoModelForTokenClassification.from_pretrained("bert-base-uncased-swa")
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/roberta-base-kin | arnolfokam | 2021-11-24T11:46:30Z | 6 | 0 | transformers | [
"transformers",
"pytorch",
"roberta",
"token-classification",
"NER",
"kin",
"dataset:masakhaner",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:05Z | ---
language:
- kin
tags:
- NER
datasets:
- masakhaner
metrics:
- f1
- precision
- recall
license: apache-2.0
widget:
- text: "Ambasaderi Bellomo yavuze ko bishimira ubufatanye burambye hagati ya EU nโu Rwanda, bushingiye nanone ku bufatanye hagati yโimigabane ya Afurika nโu Burayi."
---
# Model description
**roberta-base-kin** is a model based on the fine-tuned RoBERTa base 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 Kinyarwanda corpus **(kin)** 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 Kinyarwandan corpus **(kin)** 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
-|-|-|-
**roberta-base-kin**| 76.26 | 80.58 |78.36
# Usage
```python
from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
tokenizer = AutoTokenizer.from_pretrained("arnolfokam/roberta-base-kin")
model = AutoModelForTokenClassification.from_pretrained("arnolfokam/roberta-base-kin")
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
example = "Rayon Sports yasinyishije rutahizamu wโUmurundi"
ner_results = nlp(example)
print(ner_results)
``` |
arnolfokam/roberta-base-swa | arnolfokam | 2021-11-24T11:41:03Z | 13 | 0 | transformers | [
"transformers",
"pytorch",
"roberta",
"token-classification",
"NER",
"swa",
"dataset:masakhaner",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:05Z | ---
language:
- swa
tags:
- NER
datasets:
- masakhaner
metrics:
- f1
- precision
- recall
license: apache-2.0
widget:
- text: "Wizara ya afya ya Tanzania imeripoti Jumatatu kuwa, watu takriban 14 zaidi wamepata maambukizi ya Covid-19."
---
# Model description
**roberta-base-swa** is a model based on the fine-tuned RoBERTa base 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 Swahili corpus **(swa)** 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 **(swa)** 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
-|-|-|-
**roberta-base-swa**| 80.58 | 86.79 | 83.57
# Usage
```python
from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
tokenizer = AutoTokenizer.from_pretrained("arnolfokam/roberta-base-swa")
model = AutoModelForTokenClassification.from_pretrained("arnolfokam/roberta-base-swa")
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-ner-swa | arnolfokam | 2021-11-24T11:31:30Z | 11 | 0 | transformers | [
"transformers",
"pytorch",
"bert",
"token-classification",
"NER",
"swa",
"dataset:masakhaner",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:05Z | ---
language:
- swa
tags:
- NER
datasets:
- masakhaner
metrics:
- f1
- precision
- recall
license: apache-2.0
widget:
- text: "Wizara ya afya ya Tanzania imeripoti Jumatatu kuwa, watu takriban 14 zaidi wamepata maambukizi ya Covid-19."
---
# Model description
**mbert-base-uncased-ner-swa** is a model based on the fine-tuned Multilingual BERT base uncased model, previously fine-tuned for Named Entity Recognition using 10 high-resourced languages. 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 Swahili corpus **(swa)** 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 **(swa)** 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-ner-swa**| 82.85 | 88.13 | 85.41
# Usage
```python
from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
tokenizer = AutoTokenizer.from_pretrained("arnolfokam/mbert-base-uncased-ner-swa")
model = AutoModelForTokenClassification.from_pretrained("arnolfokam/mbert-base-uncased-ner-swa")
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-kin | arnolfokam | 2021-11-24T11:13:53Z | 10 | 0 | transformers | [
"transformers",
"pytorch",
"bert",
"token-classification",
"NER",
"kin",
"dataset:masakhaner",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:05Z | ---
language:
- kin
tags:
- NER
datasets:
- masakhaner
metrics:
- f1
- precision
- recall
license: apache-2.0
widget:
- text: "Ambasaderi Bellomo yavuze ko bishimira ubufatanye burambye hagati ya EU nโu Rwanda, bushingiye nanone ku bufatanye hagati yโimigabane ya Afurika nโu Burayi."
---
# Model description
**mbert-base-uncased-kin** 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 Kinyarwanda corpus **(kin)** 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 Kinyarwandan corpus **(kin)** 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-kin**| 81.35 | 83.98 | 82.64
# Usage
```python
from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
tokenizer = AutoTokenizer.from_pretrained("arnolfokam/mbert-base-uncased-kin")
model = AutoModelForTokenClassification.from_pretrained("arnolfokam/mbert-base-uncased-kin")
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
example = "Rayon Sports yasinyishije rutahizamu wโUmurundi"
ner_results = nlp(example)
print(ner_results)
``` |
Peterard/distilbert_feature_classifier | Peterard | 2021-11-24T03:59:16Z | 5 | 0 | transformers | [
"transformers",
"pytorch",
"distilbert",
"text-classification",
"en",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text-classification | 2022-03-02T23:29:04Z | ---
language:
- en
tags:
- text-classification
widget:
- text: "Please add a like button!"
example_title: "Likely feature request"
- text: "The app crashed when I opened it this morning. Can you fix this please?"
example_title: "Unlikely feature request"
---
How to use this classifier:
```
from transformers import pipeline
pipe = pipeline("text-classification", model="Peterard/distilbert_feature_classifier")
pipe("Please add a like button!")
# [{'label': 'feature_request', 'score': 0.8930749893188477}]
pipe("The app crashed when I opened it this morning. Can you fix this please?")
#[{'label': 'no_feature_request', 'score': 0.9971746206283569}]
```
N.B. The label will change depending on which is the likelier class |
ueb1/IceBERT-finetuned-grouped | ueb1 | 2021-11-24T00:18:29Z | 4 | 0 | transformers | [
"transformers",
"pytorch",
"tensorboard",
"roberta",
"text-classification",
"generated_from_trainer",
"license:gpl-3.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text-classification | 2022-03-02T23:29:05Z | ---
license: gpl-3.0
tags:
- generated_from_trainer
metrics:
- accuracy
model-index:
- name: IceBERT-finetuned-grouped
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. -->
# IceBERT-finetuned-grouped
This model is a fine-tuned version of [vesteinn/IceBERT](https://huggingface.co/vesteinn/IceBERT) on an unknown dataset.
It achieves the following results on the evaluation set:
- Loss: 3.5660
- Accuracy: 0.2259
## 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: 5
### Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|:-------------:|:-----:|:----:|:---------------:|:--------:|
| No log | 1.0 | 269 | 4.1727 | 0.1172 |
| 4.3535 | 2.0 | 538 | 3.8406 | 0.1632 |
| 4.3535 | 3.0 | 807 | 3.6718 | 0.2113 |
| 3.6711 | 4.0 | 1076 | 3.5660 | 0.2259 |
| 3.6711 | 5.0 | 1345 | 3.5332 | 0.2176 |
### Framework versions
- Transformers 4.12.5
- Pytorch 1.10.0+cu111
- Datasets 1.15.1
- Tokenizers 0.10.3
|
Hellisotherpeople/debate2vec | Hellisotherpeople | 2021-11-23T18:45:27Z | 34 | 7 | fasttext | [
"fasttext",
"text-classification",
"region:us"
] | text-classification | 2022-03-02T23:29:04Z | ---
tags:
- text-classification
library_name: fasttext
widget:
- text: "dialectics"
example_title: "dialectics"
- text: "schizoanalysis"
example_title: "schizoanalysis"
- text: "praxis"
example_title: "praxis"
- text: "topicality"
example_title: "topicality"
---
# debate2vec
Word-vectors created from a large corpus of competitive debate evidence, and data extraction / processing scripts
#usage
```
import fasttext.util
ft = fasttext.load_model('debate2vec.bin')
ft.get_word_vector('dialectics')
```
# Download Link
Github won't let me store large files in their repos.
* [FastText Vectors Here](https://drive.google.com/file/d/1m-CwPcaIUun4qvg69Hx2gom9dMScuQwS/view?usp=sharing) (~260mb)
# About
Created from all publically available Cross Examination Competitive debate evidence posted by the community on [Open Evidence](https://openev.debatecoaches.org/) (From 2013-2020)
Search through the original evidence by going to [debate.cards](http://debate.cards/)
Stats about this corpus:
* 222485 unique documents larger than 200 words (DebateSum plus some additional debate docs that weren't well-formed enough for inclusion into DebateSum)
* 107555 unique words (showing up more than 10 times in the corpus)
* 101 million total words
Stats about debate2vec vectors:
* 300 dimensions, minimum number of appearances of a word was 10, trained for 100 epochs with lr set to 0.10 using FastText
* lowercased (will release cased)
* No subword information
The corpus includes the following topics
* 2013-2014 Cuba/Mexico/Venezuela Economic Engagement
* 2014-2015 Oceans
* 2015-2016 Domestic Surveillance
* 2016-2017 China
* 2017-2018 Education
* 2018-2019 Immigration
* 2019-2020 Reducing Arms Sales
Other topics that this word vector model will handle extremely well
* Philosophy (Especially Left-Wing / Post-modernist)
* Law
* Government
* Politics
Initial release is of fasttext vectors without subword information. Future releases will include fine-tuned GPT-2 and other high end models as my GPU compute allows.
# Screenshots
|
AryanLala/autonlp-Scientific_Title_Generator-34558227 | AryanLala | 2021-11-23T16:51:34Z | 8 | 19 | transformers | [
"transformers",
"pytorch",
"pegasus",
"text2text-generation",
"autonlp",
"en",
"dataset:AryanLala/autonlp-data-Scientific_Title_Generator",
"co2_eq_emissions",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text2text-generation | 2022-03-02T23:29:04Z | ---
tags: autonlp
language: en
widget:
- text: "The scale, variety, and quantity of publicly-available NLP datasets has grown rapidly as researchers propose new tasks, larger models, and novel benchmarks. Datasets is a community library for contemporary NLP designed to support this ecosystem. Datasets aims to standardize end-user interfaces, versioning, and documentation, while providing a lightweight front-end that behaves similarly for small datasets as for internet-scale corpora. The design of the library incorporates a distributed, community-driven approach to adding datasets and documenting usage. After a year of development, the library now includes more than 650 unique datasets, has more than 250 contributors, and has helped support a variety of novel cross-dataset research projects and shared tasks. The library is available at https://github.com/huggingface/datasets."
datasets:
- AryanLala/autonlp-data-Scientific_Title_Generator
co2_eq_emissions: 137.60574081887984
---
# Model Trained Using AutoNLP
- Model: Google's Pegasus (https://huggingface.co/google/pegasus-xsum)
- Problem type: Summarization
- Model ID: 34558227
- CO2 Emissions (in grams): 137.60574081887984
- Spaces: https://huggingface.co/spaces/TitleGenerators/ArxivTitleGenerator
- Dataset: arXiv Dataset (https://www.kaggle.com/Cornell-University/arxiv)
- Data subset used: https://huggingface.co/datasets/AryanLala/autonlp-data-Scientific_Title_Generator
## Validation Metrics
- Loss: 2.578599214553833
- Rouge1: 44.8482
- Rouge2: 24.4052
- RougeL: 40.1716
- RougeLsum: 40.1396
- Gen Len: 11.4675
## Social
- LinkedIn: https://www.linkedin.com/in/aryanlala/
- Twitter: https://twitter.com/AryanLala20
## 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/AryanLala/autonlp-Scientific_Title_Generator-34558227
``` |
Bharathdamu/wav2vec2-large-xls-r-300m-hindi-colab | Bharathdamu | 2021-11-23T09:32:23Z | 5 | 0 | transformers | [
"transformers",
"pytorch",
"tensorboard",
"wav2vec2",
"automatic-speech-recognition",
"generated_from_trainer",
"dataset:common_voice",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | automatic-speech-recognition | 2022-03-02T23:29:04Z | ---
license: apache-2.0
tags:
- generated_from_trainer
datasets:
- common_voice
model-index:
- name: wav2vec2-large-xls-r-300m-hindi-colab
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-large-xls-r-300m-hindi-colab
This model is a fine-tuned version of [facebook/wav2vec2-xls-r-300m](https://huggingface.co/facebook/wav2vec2-xls-r-300m) on the common_voice 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: 0.0003
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 32
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 30
- mixed_precision_training: Native AMP
### Training results
### Framework versions
- Transformers 4.11.3
- Pytorch 1.10.0+cu111
- Datasets 1.13.3
- Tokenizers 0.10.3
|
Aimendo/autonlp-triage-35248482 | Aimendo | 2021-11-23T08:03:14Z | 8 | 0 | transformers | [
"transformers",
"pytorch",
"bert",
"text-classification",
"autonlp",
"en",
"dataset:Aimendo/autonlp-data-triage",
"co2_eq_emissions",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text-classification | 2022-03-02T23:29:04Z | ---
tags: autonlp
language: en
widget:
- text: "I love AutoNLP ๐ค"
datasets:
- Aimendo/autonlp-data-triage
co2_eq_emissions: 7.989144645413398
---
# Model Trained Using AutoNLP
- Problem type: Multi-class Classification
- Model ID: 35248482
- CO2 Emissions (in grams): 7.989144645413398
## Validation Metrics
- Loss: 0.13783401250839233
- Accuracy: 0.9728654124457308
- Macro F1: 0.949537871674076
- Micro F1: 0.9728654124457308
- Weighted F1: 0.9732422812610365
- Macro Precision: 0.9380372699332605
- Micro Precision: 0.9728654124457308
- Weighted Precision: 0.974548513256663
- Macro Recall: 0.9689346153591594
- Micro Recall: 0.9728654124457308
- Weighted Recall: 0.9728654124457308
## 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/Aimendo/autonlp-triage-35248482
```
Or Python API:
```
from transformers import AutoModelForSequenceClassification, AutoTokenizer
model = AutoModelForSequenceClassification.from_pretrained("Aimendo/autonlp-triage-35248482", use_auth_token=True)
tokenizer = AutoTokenizer.from_pretrained("Aimendo/autonlp-triage-35248482", use_auth_token=True)
inputs = tokenizer("I love AutoNLP", return_tensors="pt")
outputs = model(**inputs)
``` |
DeepPavlov/rubert-base-cased | DeepPavlov | 2021-11-23T08:03:04Z | 205,575 | 95 | transformers | [
"transformers",
"pytorch",
"jax",
"bert",
"feature-extraction",
"ru",
"arxiv:1905.07213",
"endpoints_compatible",
"region:us"
] | feature-extraction | 2022-03-02T23:29:04Z | ---
language:
- ru
---
# rubert-base-cased
RuBERT \(Russian, cased, 12โlayer, 768โhidden, 12โheads, 180M parameters\) was trained on the Russian part of Wikipedia and news data. We used this training data to build a vocabulary of Russian subtokens and took a multilingual version of BERTโbase as an initialization for RuBERT\[1\].
08.11.2021: upload model with MLM and NSP heads
\[1\]: Kuratov, Y., Arkhipov, M. \(2019\). Adaptation of Deep Bidirectional Multilingual Transformers for Russian Language. arXiv preprint [arXiv:1905.07213](https://arxiv.org/abs/1905.07213).
|
Maltehb/aelaectra-danish-electra-small-uncased | Maltehb | 2021-11-23T06:39:20Z | 16 | 0 | transformers | [
"transformers",
"pytorch",
"electra",
"pretraining",
"รฆlรฆctra",
"danish",
"ELECTRA-Small",
"replaced token detection",
"da",
"dataset:DAGW",
"arxiv:2003.10555",
"arxiv:1810.04805",
"arxiv:2005.03521",
"license:mit",
"co2_eq_emissions",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04Z | ---
language: "da"
co2_eq_emissions: 4009.5
tags:
- รฆlรฆctra
- pytorch
- danish
- ELECTRA-Small
- replaced token detection
license: "mit"
datasets:
- DAGW
metrics:
- f1
---
# รlรฆctra - A Step Towards More Efficient Danish Natural Language Processing
**รlรฆctra** is a Danish Transformer-based language model created to enhance the variety of Danish NLP resources with a more efficient model compared to previous state-of-the-art (SOTA) models. Initially a cased and an uncased model are released. It was created as part of a Cognitive Science bachelor's thesis.
รlรฆctra was pretrained with the ELECTRA-Small (Clark et al., 2020) pretraining approach by using the Danish Gigaword Corpus (Strรธmberg-Derczynski et al., 2020) and evaluated on Named Entity Recognition (NER) tasks. Since NER only presents a limited picture of รlรฆctra's capabilities I am very interested in further evaluations. Therefore, if you employ it for any task, feel free to hit me up your findings!
รlรฆctra was, as mentioned, created to enhance the Danish NLP capabilties and please do note how this GitHub still does not support the Danish characters "*ร, ร and ร
*" as the title of this repository becomes "*-l-ctra*". How ironic.๐
Here is an example on how to load both the cased and the uncased รlรฆctra model in [PyTorch](https://pytorch.org/) using the [๐คTransformers](https://github.com/huggingface/transformers) library:
```python
from transformers import AutoTokenizer, AutoModelForPreTraining
tokenizer = AutoTokenizer.from_pretrained("Maltehb/-l-ctra-cased")
model = AutoModelForPreTraining.from_pretrained("Maltehb/-l-ctra-cased")
```
```python
from transformers import AutoTokenizer, AutoModelForPreTraining
tokenizer = AutoTokenizer.from_pretrained("Maltehb/-l-ctra-uncased")
model = AutoModelForPreTraining.from_pretrained("Maltehb/-l-ctra-uncased")
```
### Evaluation of current Danish Language Models
รlรฆctra, Danish BERT (DaBERT) and multilingual BERT (mBERT) were evaluated:
| Model | Layers | Hidden Size | Params | AVG NER micro-f1 (DaNE-testset) | Average Inference Time (Sec/Epoch) | Download |
| --- | --- | --- | --- | --- | --- | --- |
| รlรฆctra Uncased | 12 | 256 | 13.7M | 78.03 (SD = 1.28) | 10.91 | [Link for model](https://www.dropbox.com/s/cag7prs1nvdchqs/%C3%86l%C3%A6ctra.zip?dl=0) |
| รlรฆctra Cased | 12 | 256 | 14.7M | 80.08 (SD = 0.26) | 10.92 | [Link for model](https://www.dropbox.com/s/cag7prs1nvdchqs/%C3%86l%C3%A6ctra.zip?dl=0) |
| DaBERT | 12 | 768 | 110M | 84.89 (SD = 0.64) | 43.03 | [Link for model](https://www.dropbox.com/s/19cjaoqvv2jicq9/danish_bert_uncased_v2.zip?dl=1) |
| mBERT Uncased | 12 | 768 | 167M | 80.44 (SD = 0.82) | 72.10 | [Link for model](https://storage.googleapis.com/bert_models/2018_11_03/multilingual_L-12_H-768_A-12.zip) |
| mBERT Cased | 12 | 768 | 177M | 83.79 (SD = 0.91) | 70.56 | [Link for model](https://storage.googleapis.com/bert_models/2018_11_23/multi_cased_L-12_H-768_A-12.zip) |
On [DaNE](https://danlp.alexandra.dk/304bd159d5de/datasets/ddt.zip) (Hvingelby et al., 2020), รlรฆctra scores slightly worse than both cased and uncased Multilingual BERT (Devlin et al., 2019) and Danish BERT (Danish BERT, 2019/2020), however, รlรฆctra is less than one third the size, and uses significantly fewer computational resources to pretrain and instantiate. For a full description of the evaluation and specification of the model read the thesis: 'รlรฆctra - A Step Towards More Efficient Danish Natural Language Processing'.
### Pretraining
To pretrain รlรฆctra it is recommended to build a Docker Container from the [Dockerfile](https://github.com/MalteHB/รlรฆctra/tree/master/notebooks/fine-tuning/). Next, simply follow the [pretraining notebooks](https://github.com/MalteHB/รlรฆctra/tree/master/infrastructure/Dockerfile/)
The pretraining was done by utilizing a single NVIDIA Tesla V100 GPU with 16 GiB, endowed by the Danish data company [KMD](https://www.kmd.dk/). The pretraining took approximately 4 days and 9.5 hours for both the cased and uncased model
### Fine-tuning
To fine-tune any รlรฆctra model follow the [fine-tuning notebooks](https://github.com/MalteHB/รlรฆctra/tree/master/notebooks/fine-tuning/)
### References
Clark, K., Luong, M.-T., Le, Q. V., & Manning, C. D. (2020). ELECTRA: Pre-training Text Encoders as Discriminators Rather Than Generators. ArXiv:2003.10555 [Cs]. http://arxiv.org/abs/2003.10555
Danish BERT. (2020). BotXO. https://github.com/botxo/nordic_bert (Original work published 2019)
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2019). BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. ArXiv:1810.04805 [Cs]. http://arxiv.org/abs/1810.04805
Hvingelby, R., Pauli, A. B., Barrett, M., Rosted, C., Lidegaard, L. M., & Sรธgaard, A. (2020). DaNE: A Named Entity Resource for Danish. Proceedings of the 12th Language Resources and Evaluation Conference, 4597โ4604. https://www.aclweb.org/anthology/2020.lrec-1.565
Strรธmberg-Derczynski, L., Baglini, R., Christiansen, M. H., Ciosici, M. R., Dalsgaard, J. A., Fusaroli, R., Henrichsen, P. J., Hvingelby, R., Kirkedal, A., Kjeldsen, A. S., Ladefoged, C., Nielsen, F. ร
., Petersen, M. L., Rystrรธm, J. H., & Varab, D. (2020). The Danish Gigaword Project. ArXiv:2005.03521 [Cs]. http://arxiv.org/abs/2005.03521
#### Acknowledgements
As the majority of this repository is build upon [the works](https://github.com/google-research/electra) by the team at Google who created ELECTRA, a HUGE thanks to them is in order.
A Giga thanks also goes out to the incredible people who collected The Danish Gigaword Corpus (Strรธmberg-Derczynski et al., 2020).
Furthermore, I would like to thank my supervisor [Riccardo Fusaroli](https://github.com/fusaroli) for the support with the thesis, and a special thanks goes out to [Kenneth Enevoldsen](https://github.com/KennethEnevoldsen) for his continuous feedback.
Lastly, i would like to thank KMD, my colleagues from KMD, and my peers and co-students from Cognitive Science for encouriging me to keep on working hard and holding my head up high!
#### Contact
For help or further information feel free to connect with the author Malte Hรธjmark-Bertelsen on [[email protected]](mailto:[email protected]?subject=[GitHub]%20รlรฆctra) or any of the following platforms:
[<img align="left" alt="MalteHB | Twitter" width="22px" src="https://cdn.jsdelivr.net/npm/simple-icons@v3/icons/twitter.svg" />][twitter]
[<img align="left" alt="MalteHB | LinkedIn" width="22px" src="https://cdn.jsdelivr.net/npm/simple-icons@v3/icons/linkedin.svg" />][linkedin]
[<img align="left" alt="MalteHB | Instagram" width="22px" src="https://cdn.jsdelivr.net/npm/simple-icons@v3/icons/instagram.svg" />][instagram]
<br />
</details>
[twitter]: https://twitter.com/malteH_B
[instagram]: https://www.instagram.com/maltemusen/
[linkedin]: https://www.linkedin.com/in/malte-h%C3%B8jmark-bertelsen-9a618017b/ |
artursz/wav2vec2-large-xls-r-300m-lv-v05 | artursz | 2021-11-23T02:47:04Z | 4 | 0 | transformers | [
"transformers",
"pytorch",
"tensorboard",
"wav2vec2",
"automatic-speech-recognition",
"generated_from_trainer",
"dataset:common_voice",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | automatic-speech-recognition | 2022-03-02T23:29:05Z | ---
license: apache-2.0
tags:
- generated_from_trainer
datasets:
- common_voice
model-index:
- name: wav2vec2-large-xls-r-300m-lv-v05
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-large-xls-r-300m-lv-v05
This model is a fine-tuned version of [facebook/wav2vec2-xls-r-300m](https://huggingface.co/facebook/wav2vec2-xls-r-300m) on the common_voice dataset.
It achieves the following results on the evaluation set:
- Loss: 0.3862
- Wer: 0.2588
## 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: 0.0003
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 32
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 50
- mixed_precision_training: Native AMP
### Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|:-------------:|:-----:|:----:|:---------------:|:------:|
| 4.8836 | 2.81 | 400 | 0.8722 | 0.7244 |
| 0.5365 | 5.63 | 800 | 0.4622 | 0.4812 |
| 0.277 | 8.45 | 1200 | 0.4348 | 0.4056 |
| 0.1947 | 11.27 | 1600 | 0.4223 | 0.3636 |
| 0.1655 | 14.08 | 2000 | 0.4084 | 0.3465 |
| 0.1441 | 16.9 | 2400 | 0.4329 | 0.3497 |
| 0.121 | 19.72 | 2800 | 0.4371 | 0.3324 |
| 0.1062 | 22.53 | 3200 | 0.4202 | 0.3198 |
| 0.0937 | 25.35 | 3600 | 0.4063 | 0.3265 |
| 0.0871 | 28.17 | 4000 | 0.4253 | 0.3255 |
| 0.0755 | 30.98 | 4400 | 0.4368 | 0.3194 |
| 0.0627 | 33.8 | 4800 | 0.4067 | 0.2908 |
| 0.0595 | 36.62 | 5200 | 0.3929 | 0.2973 |
| 0.0523 | 39.44 | 5600 | 0.3748 | 0.2817 |
| 0.0434 | 42.25 | 6000 | 0.3769 | 0.2711 |
| 0.0391 | 45.07 | 6400 | 0.3901 | 0.2653 |
| 0.0319 | 47.88 | 6800 | 0.3862 | 0.2588 |
### Framework versions
- Transformers 4.11.3
- Pytorch 1.10.0+cu111
- Datasets 1.13.3
- Tokenizers 0.10.3
|
huggingtweets/kylelchong | huggingtweets | 2021-11-23T01:12:59Z | 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: https://www.huggingtweets.com/kylelchong/1637629975064/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/1363977743021584394/17Z8FHm2_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">๐ค AI BOT ๐ค</div>
<div style="text-align: center; font-size: 16px; font-weight: 800">Kyle L. Chong (he.him.his)</div>
<div style="text-align: center; font-size: 14px;">@kylelchong</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 Kyle L. Chong (he.him.his).
| Data | Kyle L. Chong (he.him.his) |
| --- | --- |
| Tweets downloaded | 1072 |
| Retweets | 213 |
| Short tweets | 76 |
| Tweets kept | 783 |
[Explore the data](https://wandb.ai/wandb/huggingtweets/runs/2xlb7d6c/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 @kylelchong's tweets.
Hyperparameters and metrics are recorded in the [W&B training run](https://wandb.ai/wandb/huggingtweets/runs/5bvgy2zz) for full transparency and reproducibility.
At the end of training, [the final model](https://wandb.ai/wandb/huggingtweets/runs/5bvgy2zz/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/kylelchong')
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)
|
gayanin/t5-small-mlm-pubmed-35 | gayanin | 2021-11-22T22:24:30Z | 4 | 0 | transformers | [
"transformers",
"pytorch",
"tensorboard",
"t5",
"text2text-generation",
"generated_from_trainer",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text2text-generation | 2022-03-02T23:29:05Z | ---
license: apache-2.0
tags:
- generated_from_trainer
model-index:
- name: t5-small-mlm-pubmed-35
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-small-mlm-pubmed-35
This model is a fine-tuned version of [t5-small](https://huggingface.co/t5-small) on an unknown dataset.
It achieves the following results on the evaluation set:
- Loss: 1.1101
- Rouge2 Precision: 0.4758
- Rouge2 Recall: 0.3498
- Rouge2 Fmeasure: 0.3927
## 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: 40
- mixed_precision_training: Native AMP
### Training results
| Training Loss | Epoch | Step | Validation Loss | Rouge2 Precision | Rouge2 Recall | Rouge2 Fmeasure |
|:-------------:|:-----:|:-----:|:---------------:|:----------------:|:-------------:|:---------------:|
| 1.8404 | 0.75 | 500 | 1.5005 | 0.4265 | 0.2786 | 0.3273 |
| 1.6858 | 1.51 | 1000 | 1.4216 | 0.4318 | 0.2946 | 0.3404 |
| 1.6071 | 2.26 | 1500 | 1.3777 | 0.4472 | 0.3148 | 0.3598 |
| 1.5551 | 3.02 | 2000 | 1.3360 | 0.4406 | 0.3168 | 0.3586 |
| 1.5116 | 3.77 | 2500 | 1.3128 | 0.4523 | 0.3234 | 0.3671 |
| 1.4837 | 4.52 | 3000 | 1.2937 | 0.4477 | 0.3215 | 0.3645 |
| 1.4513 | 5.28 | 3500 | 1.2766 | 0.4511 | 0.3262 | 0.3689 |
| 1.4336 | 6.03 | 4000 | 1.2626 | 0.4548 | 0.3283 | 0.3718 |
| 1.4149 | 6.79 | 4500 | 1.2449 | 0.4495 | 0.3274 | 0.3687 |
| 1.3977 | 7.54 | 5000 | 1.2349 | 0.4507 | 0.3305 | 0.3712 |
| 1.3763 | 8.3 | 5500 | 1.2239 | 0.4519 | 0.3266 | 0.3688 |
| 1.371 | 9.05 | 6000 | 1.2171 | 0.4546 | 0.3305 | 0.3727 |
| 1.3501 | 9.8 | 6500 | 1.2080 | 0.4575 | 0.3329 | 0.3755 |
| 1.3443 | 10.56 | 7000 | 1.2017 | 0.4576 | 0.3314 | 0.3742 |
| 1.326 | 11.31 | 7500 | 1.1926 | 0.4578 | 0.333 | 0.3757 |
| 1.3231 | 12.07 | 8000 | 1.1866 | 0.4606 | 0.3357 | 0.3782 |
| 1.3089 | 12.82 | 8500 | 1.1816 | 0.4591 | 0.3338 | 0.3765 |
| 1.3007 | 13.57 | 9000 | 1.1764 | 0.4589 | 0.3361 | 0.3777 |
| 1.2943 | 14.33 | 9500 | 1.1717 | 0.4641 | 0.3382 | 0.3811 |
| 1.2854 | 15.08 | 10000 | 1.1655 | 0.4617 | 0.3378 | 0.38 |
| 1.2777 | 15.84 | 10500 | 1.1612 | 0.464 | 0.3401 | 0.3823 |
| 1.2684 | 16.59 | 11000 | 1.1581 | 0.4608 | 0.3367 | 0.3789 |
| 1.2612 | 17.35 | 11500 | 1.1554 | 0.4623 | 0.3402 | 0.3818 |
| 1.2625 | 18.1 | 12000 | 1.1497 | 0.4613 | 0.3381 | 0.3802 |
| 1.2529 | 18.85 | 12500 | 1.1465 | 0.4671 | 0.3419 | 0.3848 |
| 1.2461 | 19.61 | 13000 | 1.1431 | 0.4646 | 0.3399 | 0.3824 |
| 1.2415 | 20.36 | 13500 | 1.1419 | 0.4659 | 0.341 | 0.3835 |
| 1.2375 | 21.12 | 14000 | 1.1377 | 0.4693 | 0.3447 | 0.3873 |
| 1.2315 | 21.87 | 14500 | 1.1353 | 0.4672 | 0.3433 | 0.3855 |
| 1.2263 | 22.62 | 15000 | 1.1333 | 0.467 | 0.3433 | 0.3854 |
| 1.2214 | 23.38 | 15500 | 1.1305 | 0.4682 | 0.3446 | 0.3869 |
| 1.2202 | 24.13 | 16000 | 1.1291 | 0.4703 | 0.3465 | 0.3888 |
| 1.2155 | 24.89 | 16500 | 1.1270 | 0.472 | 0.348 | 0.3903 |
| 1.2064 | 25.64 | 17000 | 1.1261 | 0.4724 | 0.3479 | 0.3905 |
| 1.2173 | 26.4 | 17500 | 1.1236 | 0.4734 | 0.3485 | 0.3912 |
| 1.1994 | 27.15 | 18000 | 1.1220 | 0.4739 | 0.3486 | 0.3915 |
| 1.2018 | 27.9 | 18500 | 1.1217 | 0.4747 | 0.3489 | 0.3921 |
| 1.2045 | 28.66 | 19000 | 1.1194 | 0.4735 | 0.3488 | 0.3916 |
| 1.1949 | 29.41 | 19500 | 1.1182 | 0.4732 | 0.3484 | 0.3911 |
| 1.19 | 30.17 | 20000 | 1.1166 | 0.4724 | 0.3479 | 0.3904 |
| 1.1932 | 30.92 | 20500 | 1.1164 | 0.4753 | 0.3494 | 0.3924 |
| 1.1952 | 31.67 | 21000 | 1.1147 | 0.4733 | 0.3485 | 0.3911 |
| 1.1922 | 32.43 | 21500 | 1.1146 | 0.475 | 0.3494 | 0.3923 |
| 1.1889 | 33.18 | 22000 | 1.1132 | 0.4765 | 0.3499 | 0.3933 |
| 1.1836 | 33.94 | 22500 | 1.1131 | 0.4768 | 0.351 | 0.3939 |
| 1.191 | 34.69 | 23000 | 1.1127 | 0.4755 | 0.3495 | 0.3926 |
| 1.1811 | 35.44 | 23500 | 1.1113 | 0.4748 | 0.349 | 0.3919 |
| 1.1864 | 36.2 | 24000 | 1.1107 | 0.4751 | 0.3494 | 0.3921 |
| 1.1789 | 36.95 | 24500 | 1.1103 | 0.4756 | 0.3499 | 0.3927 |
| 1.1819 | 37.71 | 25000 | 1.1101 | 0.4758 | 0.35 | 0.3932 |
| 1.1862 | 38.46 | 25500 | 1.1099 | 0.4755 | 0.3497 | 0.3926 |
| 1.1764 | 39.22 | 26000 | 1.1101 | 0.4759 | 0.3498 | 0.3928 |
| 1.1819 | 39.97 | 26500 | 1.1101 | 0.4758 | 0.3498 | 0.3927 |
### Framework versions
- Transformers 4.12.5
- Pytorch 1.10.0+cu111
- Datasets 1.15.1
- Tokenizers 0.10.3
|
gayanin/bart-mlm-pubmed-35 | gayanin | 2021-11-22T21:16:10Z | 5 | 0 | transformers | [
"transformers",
"pytorch",
"tensorboard",
"bart",
"text2text-generation",
"generated_from_trainer",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text2text-generation | 2022-03-02T23:29:05Z | ---
license: apache-2.0
tags:
- generated_from_trainer
model-index:
- name: bart-mlm-pubmed-35
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. -->
# bart-mlm-pubmed-35
This model is a fine-tuned version of [facebook/bart-base](https://huggingface.co/facebook/bart-base) on an unknown dataset.
It achieves the following results on the evaluation set:
- Loss: 0.9359
- Rouge2 Precision: 0.5451
- Rouge2 Recall: 0.4232
- Rouge2 Fmeasure: 0.4666
## 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: 10
- mixed_precision_training: Native AMP
### Training results
| Training Loss | Epoch | Step | Validation Loss | Rouge2 Precision | Rouge2 Recall | Rouge2 Fmeasure |
|:-------------:|:-----:|:----:|:---------------:|:----------------:|:-------------:|:---------------:|
| 1.4156 | 1.0 | 663 | 1.0366 | 0.5165 | 0.3967 | 0.4394 |
| 1.1773 | 2.0 | 1326 | 0.9841 | 0.5354 | 0.4168 | 0.4589 |
| 1.0894 | 3.0 | 1989 | 0.9554 | 0.5346 | 0.4133 | 0.4563 |
| 0.9359 | 4.0 | 2652 | 0.9440 | 0.5357 | 0.4163 | 0.4587 |
| 0.8758 | 5.0 | 3315 | 0.9340 | 0.5428 | 0.4226 | 0.465 |
| 0.8549 | 6.0 | 3978 | 0.9337 | 0.5385 | 0.422 | 0.4634 |
| 0.7743 | 7.0 | 4641 | 0.9330 | 0.542 | 0.422 | 0.4647 |
| 0.7465 | 8.0 | 5304 | 0.9315 | 0.5428 | 0.4231 | 0.4654 |
| 0.7348 | 9.0 | 5967 | 0.9344 | 0.5462 | 0.4244 | 0.4674 |
| 0.7062 | 10.0 | 6630 | 0.9359 | 0.5451 | 0.4232 | 0.4666 |
### Framework versions
- Transformers 4.12.5
- Pytorch 1.10.0+cu111
- Datasets 1.15.1
- Tokenizers 0.10.3
|
gayanin/bart-mlm-pubmed-15 | gayanin | 2021-11-22T20:33:06Z | 6 | 0 | transformers | [
"transformers",
"pytorch",
"tensorboard",
"bart",
"text2text-generation",
"generated_from_trainer",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text2text-generation | 2022-03-02T23:29:05Z | ---
license: apache-2.0
tags:
- generated_from_trainer
model-index:
- name: bart-mlm-pubmed-15
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. -->
# bart-mlm-pubmed-15
This model is a fine-tuned version of [facebook/bart-base](https://huggingface.co/facebook/bart-base) on an unknown dataset.
It achieves the following results on the evaluation set:
- Loss: 0.4822
- Rouge2 Precision: 0.7578
- Rouge2 Recall: 0.5933
- Rouge2 Fmeasure: 0.6511
## 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: 10
- mixed_precision_training: Native AMP
### Training results
| Training Loss | Epoch | Step | Validation Loss | Rouge2 Precision | Rouge2 Recall | Rouge2 Fmeasure |
|:-------------:|:-----:|:----:|:---------------:|:----------------:|:-------------:|:---------------:|
| 0.7006 | 1.0 | 663 | 0.5062 | 0.7492 | 0.5855 | 0.6434 |
| 0.5709 | 2.0 | 1326 | 0.4811 | 0.7487 | 0.5879 | 0.6447 |
| 0.5011 | 3.0 | 1989 | 0.4734 | 0.7541 | 0.5906 | 0.6483 |
| 0.4164 | 4.0 | 2652 | 0.4705 | 0.7515 | 0.5876 | 0.6452 |
| 0.3888 | 5.0 | 3315 | 0.4703 | 0.7555 | 0.5946 | 0.6515 |
| 0.3655 | 6.0 | 3978 | 0.4725 | 0.7572 | 0.5943 | 0.6516 |
| 0.319 | 7.0 | 4641 | 0.4733 | 0.7557 | 0.5911 | 0.6491 |
| 0.3089 | 8.0 | 5304 | 0.4792 | 0.7577 | 0.5936 | 0.6513 |
| 0.2907 | 9.0 | 5967 | 0.4799 | 0.7577 | 0.5931 | 0.6509 |
| 0.275 | 10.0 | 6630 | 0.4822 | 0.7578 | 0.5933 | 0.6511 |
### Framework versions
- Transformers 4.12.5
- Pytorch 1.10.0+cu111
- Datasets 1.15.1
- Tokenizers 0.10.3
|
samantharhay/wav2vec2-base-myst-demo-colab | samantharhay | 2021-11-22T18:15:21Z | 7 | 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-base-myst-demo-colab
---
<!-- 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-base-myst-demo-colab
This model is a fine-tuned version of [facebook/wav2vec2-base-960h](https://huggingface.co/facebook/wav2vec2-base-960h) on an unknown dataset.
It achieves the following results on the evaluation set:
- eval_loss: 1.3125
- eval_wer: 0.3139
- eval_runtime: 57.3226
- eval_samples_per_second: 9.996
- eval_steps_per_second: 1.256
- epoch: 18.68
- step: 17000
## 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: 1e-05
- train_batch_size: 2
- 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
|
JorisCos/VAD_Net | JorisCos | 2021-11-22T17:17:23Z | 7 | 0 | asteroid | [
"asteroid",
"pytorch",
"audio",
"VADNet",
"VAD",
"Voice Activity Detection",
"dataset:LibriVAD",
"license:cc-by-sa-4.0",
"region:us"
] | null | 2022-03-02T23:29:04Z | ---
tags:
- asteroid
- audio
- VADNet
- VAD
- Voice Activity Detection
datasets:
- LibriVAD
license: cc-by-sa-4.0
---
## Asteroid model `JorisCos/VAD_Net`
Description:
This model was trained by Joris Cosentino using the librimix recipe in [Asteroid](https://github.com/asteroid-team/asteroid).
It was trained on the `enh_single` task of the Libri1Mix dataset.
Training config:
```yml
data:
segment: 3
train_dir: /home/jcosentino/VAD_dataset/metadata/sets/train.json
valid_dir: /home/jcosentino/VAD_dataset/metadata/sets/dev.json
filterbank:
kernel_size: 16
n_filters: 512
stride: 8
main_args:
exp_dir: exp/full_not_causal_f1/
help: null
masknet:
bn_chan: 128
causal: false
hid_chan: 512
mask_act: relu
n_blocks: 3
n_repeats: 5
skip_chan: 128
optim:
lr: 0.001
optimizer: adam
weight_decay: 0.0
positional arguments: {}
training:
batch_size: 8
early_stop: true
epochs: 200
half_lr: true
num_workers: 4
```
Results:
On LibriVAD min test set :
```yml
accuracy: 0.8196149023502931,
precision: 0.8305009048356607,
recall: 0.8869202491310206,
f1_score: 0.8426184545700124
```
License notice:
This work "VAD_Net" is a derivative of [LibriSpeech ASR corpus](http://www.openslr.org/12) by Vassil Panayotov,
used under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/); of The [DNS challenge](https://github.com/microsoft/DNS-Challenge) noises, [Attribution-ShareAlike 3.0 Unported](https://creativecommons.org/licenses/by-sa/3.0/).
"VAD_Net" is licensed under [Attribution-ShareAlike 3.0 Unported](https://creativecommons.org/licenses/by-sa/3.0/) by Joris Cosentino |
huggingtweets/ctrlcreep | huggingtweets | 2021-11-22T09:35:47Z | 4 | 1 | 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/ctrlcreep/1637573720314/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/855460243152801793/cxX82P3V_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">๐ค AI BOT ๐ค</div>
<div style="text-align: center; font-size: 16px; font-weight: 800">infineot</div>
<div style="text-align: center; font-size: 14px;">@ctrlcreep</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 infineot.
| Data | infineot |
| --- | --- |
| Tweets downloaded | 3241 |
| Retweets | 171 |
| Short tweets | 51 |
| Tweets kept | 3019 |
[Explore the data](https://wandb.ai/wandb/huggingtweets/runs/26459hr9/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 @ctrlcreep's tweets.
Hyperparameters and metrics are recorded in the [W&B training run](https://wandb.ai/wandb/huggingtweets/runs/1prcdcpn) for full transparency and reproducibility.
At the end of training, [the final model](https://wandb.ai/wandb/huggingtweets/runs/1prcdcpn/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/ctrlcreep')
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)
|
khalidalt/DeBERTa-v3-large-mnli | khalidalt | 2021-11-22T08:38:23Z | 54 | 5 | transformers | [
"transformers",
"pytorch",
"deberta-v2",
"text-classification",
"zero-shot-classification",
"en",
"arxiv:2006.03654",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text-classification | 2022-03-02T23:29:05Z | ---
language:
- en
tags:
- text-classification
- zero-shot-classification
metrics:
- accuracy
widget:
- text: "The Movie have been criticized for the story. However, I think it is a great movie. [SEP] I liked the movie."
---
# DeBERTa-v3-large-mnli
## Model description
This model was trained on the Multi-Genre Natural Language Inference ( MultiNLI ) dataset, which consists of 433k sentence pairs textual entailment information.
The model used is [DeBERTa-v3-large from Microsoft](https://huggingface.co/microsoft/deberta-large). The v3 DeBERTa outperforms the result of Bert and RoBERTa in majority of NLU benchmarks by using disentangled attention and enhanced mask decoder. More information about the orginal model is on [official repository](https://github.com/microsoft/DeBERTa) and the [paper](https://arxiv.org/abs/2006.03654)
## Intended uses & limitations
#### How to use the model
```python
premise = "The Movie have been criticized for the story. However, I think it is a great movie."
hypothesis = "I liked the movie."
input = tokenizer(premise, hypothesis, truncation=True, return_tensors="pt")
output = model(input["input_ids"].to(device)) # device = "cuda:0" or "cpu"
prediction = torch.softmax(output["logits"][0], -1)
label_names = ["entailment", "neutral", "contradiction"]
print(label_names[prediction.argmax(0).tolist()])
```
### Training data
This model was trained on the MultiNLI dataset, which consists of 392K sentence textual entitlement.
### Training procedure
DeBERTa-v3-large-mnli was trained using the Hugging Face trainer with the following hyperparameters.
```
train_args = TrainingArguments(
learning_rate=2e-5,
per_device_train_batch_size=8,
per_device_eval_batch_size=8,
num_train_epochs=3,
warmup_ratio=0.06,
weight_decay=0.1,
fp16=True,
seed=42,
)
```
### BibTeX entry and citation info
Please cite the [DeBERTa paper](https://arxiv.org/abs/2006.03654) and [MultiNLI Dataset](https://cims.nyu.edu/~sbowman/multinli/paper.pdf) if you use this model and include this Huggingface hub. |
jsylee/scibert_scivocab_uncased-finetuned-ner | jsylee | 2021-11-22T03:52:41Z | 6,334 | 14 | transformers | [
"transformers",
"pytorch",
"bert",
"token-classification",
"Named Entity Recognition",
"SciBERT",
"Adverse Effect",
"Drug",
"Medical",
"en",
"dataset:ade_corpus_v2",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:05Z | ---
language:
- en
tags:
- Named Entity Recognition
- SciBERT
- Adverse Effect
- Drug
- Medical
datasets:
- ade_corpus_v2
widget:
- text: "Abortion, miscarriage or uterine hemorrhage associated with misoprostol (Cytotec), a labor-inducing drug."
example_title: "Abortion, miscarriage, ..."
- text: "Addiction to many sedatives and analgesics, such as diazepam, morphine, etc."
example_title: "Addiction to many..."
- text: "Birth defects associated with thalidomide"
example_title: "Birth defects associated..."
- text: "Bleeding of the intestine associated with aspirin therapy"
example_title: "Bleeding of the intestine..."
- text: "Cardiovascular disease associated with COX-2 inhibitors (i.e. Vioxx)"
example_title: "Cardiovascular disease..."
---
This is a SciBERT-based model fine-tuned to perform Named Entity Recognition for drug names and adverse drug effects.
This model classifies input tokens into one of five classes:
- `B-DRUG`: beginning of a drug entity
- `I-DRUG`: within a drug entity
- `B-EFFECT`: beginning of an AE entity
- `I-EFFECT`: within an AE entity
- `O`: outside either of the above entities
To get started using this model for inference, simply set up an NER `pipeline` like below:
```python
from transformers import (AutoModelForTokenClassification,
AutoTokenizer,
pipeline,
)
model_checkpoint = "jsylee/scibert_scivocab_uncased-finetuned-ner"
model = AutoModelForTokenClassification.from_pretrained(model_checkpoint, num_labels=5,
id2label={0: 'O', 1: 'B-DRUG', 2: 'I-DRUG', 3: 'B-EFFECT', 4: 'I-EFFECT'}
)
tokenizer = AutoTokenizer.from_pretrained(model_checkpoint)
model_pipeline = pipeline(task="ner", model=model, tokenizer=tokenizer)
print( model_pipeline ("Abortion, miscarriage or uterine hemorrhage associated with misoprostol (Cytotec), a labor-inducing drug."))
```
SciBERT: https://huggingface.co/allenai/scibert_scivocab_uncased
Dataset: https://huggingface.co/datasets/ade_corpus_v2
|
teven/roberta_kelm_tekgen | teven | 2021-11-22T01:04:55Z | 3 | 0 | sentence-transformers | [
"sentence-transformers",
"pytorch",
"roberta",
"feature-extraction",
"sentence-similarity",
"transformers",
"autotrain_compatible",
"text-embeddings-inference",
"endpoints_compatible",
"region:us"
] | sentence-similarity | 2022-03-02T23:29:05Z | ---
pipeline_tag: sentence-similarity
tags:
- sentence-transformers
- feature-extraction
- sentence-similarity
- transformers
---
# teven/roberta_kelm_tekgen
This is a [sentence-transformers](https://www.SBERT.net) model: It maps sentences & paragraphs to a 768 dimensional dense vector space and can be used for tasks like clustering or semantic search.
<!--- Describe your model here -->
## Usage (Sentence-Transformers)
Using this model becomes easy when you have [sentence-transformers](https://www.SBERT.net) installed:
```
pip install -U sentence-transformers
```
Then you can use the model like this:
```python
from sentence_transformers import SentenceTransformer
sentences = ["This is an example sentence", "Each sentence is converted"]
model = SentenceTransformer('teven/roberta_kelm_tekgen')
embeddings = model.encode(sentences)
print(embeddings)
```
## Usage (HuggingFace Transformers)
Without [sentence-transformers](https://www.SBERT.net), you can use the model like this: First, you pass your input through the transformer model, then you have to apply the right pooling-operation on-top of the contextualized word embeddings.
```python
from transformers import AutoTokenizer, AutoModel
import torch
#Mean Pooling - Take attention mask into account for correct averaging
def mean_pooling(model_output, attention_mask):
token_embeddings = model_output[0] #First element of model_output contains all token embeddings
input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)
# Sentences we want sentence embeddings for
sentences = ['This is an example sentence', 'Each sentence is converted']
# Load model from HuggingFace Hub
tokenizer = AutoTokenizer.from_pretrained('teven/roberta_kelm_tekgen')
model = AutoModel.from_pretrained('teven/roberta_kelm_tekgen')
# Tokenize sentences
encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors='pt')
# Compute token embeddings
with torch.no_grad():
model_output = model(**encoded_input)
# Perform pooling. In this case, mean pooling.
sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])
print("Sentence embeddings:")
print(sentence_embeddings)
```
## Evaluation Results
<!--- Describe how your model was evaluated -->
For an automated evaluation of this model, see the *Sentence Embeddings Benchmark*: [https://seb.sbert.net](https://seb.sbert.net?model_name=teven/roberta_kelm_tekgen)
## Training
The model was trained with the parameters:
**DataLoader**:
`torch.utils.data.dataloader.DataLoader` of length 976035 with parameters:
```
{'batch_size': 16, 'sampler': 'torch.utils.data.sampler.SequentialSampler', 'batch_sampler': 'torch.utils.data.sampler.BatchSampler'}
```
**Loss**:
`sentence_transformers.losses.MultipleNegativesRankingLoss.MultipleNegativesRankingLoss` with parameters:
```
{'scale': 20.0, 'similarity_fct': 'cos_sim'}
```
**DataLoader**:
`torch.utils.data.dataloader.DataLoader` of length 394379 with parameters:
```
{'batch_size': 16, 'sampler': 'torch.utils.data.sampler.SequentialSampler', 'batch_sampler': 'torch.utils.data.sampler.BatchSampler'}
```
**Loss**:
`sentence_transformers.losses.MultipleNegativesRankingLoss.MultipleNegativesRankingLoss` with parameters:
```
{'scale': 20.0, 'similarity_fct': 'cos_sim'}
```
Parameters of the fit()-Method:
```
[
{
"epochs": 1,
"evaluation_steps": 1000,
"evaluator": "sentence_transformers.evaluation.SequentialEvaluator.SequentialEvaluator",
"max_grad_norm": 1,
"optimizer_class": "<class 'transformers.optimization.AdamW'>",
"optimizer_params": {
"lr": 2e-05
},
"scheduler": "WarmupLinear",
"steps_per_epoch": null,
"warmup_steps": 1000,
"weight_decay": 0.01
}
]
```
## Full Model Architecture
```
SentenceTransformer(
(0): Transformer({'max_seq_length': 300, 'do_lower_case': False}) with Transformer model: RobertaModel
(1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False})
)
```
## Citing & Authors
<!--- Describe where people can find more information --> |
Ulto/pythonCoPilot2 | Ulto | 2021-11-22T00:24:53Z | 5 | 0 | transformers | [
"transformers",
"pytorch",
"tensorboard",
"gpt2",
"text-generation",
"generated_from_trainer",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text-generation | 2022-03-02T23:29:05Z | ---
tags:
- generated_from_trainer
model-index:
- name: pythonCoPilot2
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. -->
# pythonCoPilot2
This model is a fine-tuned version of [](https://huggingface.co/) on an unknown dataset.
It achieves the following results on the evaluation set:
- Loss: 4.0479
## 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 |
|:-------------:|:-----:|:----:|:---------------:|
| No log | 1.0 | 427 | 4.3782 |
| 4.6698 | 2.0 | 854 | 4.0718 |
| 3.3953 | 3.0 | 1281 | 4.0479 |
### Framework versions
- Transformers 4.12.5
- Pytorch 1.10.0+cu111
- Datasets 1.15.1
- Tokenizers 0.10.3
|
Ulto/pythonCoPilot | Ulto | 2021-11-21T23:49:37Z | 5 | 0 | transformers | [
"transformers",
"pytorch",
"tensorboard",
"gpt2",
"text-generation",
"generated_from_trainer",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text-generation | 2022-03-02T23:29:05Z | ---
tags:
- generated_from_trainer
model-index:
- name: pythonCoPilot
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. -->
# pythonCoPilot
This model is a fine-tuned version of [](https://huggingface.co/) 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: 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
### Framework versions
- Transformers 4.12.5
- Pytorch 1.10.0+cu111
- Datasets 1.15.1
- Tokenizers 0.10.3
|
KrishParikh/gpt2_imdb_movie_plots | KrishParikh | 2021-11-21T20:11:06Z | 5 | 1 | transformers | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"generated_from_trainer",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | text-generation | 2022-03-02T23:29:04Z | ---
tags:
- generated_from_trainer
model-index:
- name: gpt2-plot
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. -->
# gpt2-plot
This model is a fine-tuned version of [gpt2-medium](https://huggingface.co/gpt2-medium) on an unknown dataset.
It achieves the following results on the evaluation set:
- Loss: 2.8856
## 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: 1
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 5.0
- mixed_precision_training: Native AMP
### Training results
### Framework versions
- Transformers 4.13.0.dev0
- Pytorch 1.9.0
- Datasets 1.15.1
- Tokenizers 0.10.3
|
Abirate/bert_fine_tuned_cola | Abirate | 2021-11-21T16:41:00Z | 10 | 1 | transformers | [
"transformers",
"tf",
"bert",
"text-classification",
"arxiv:1810.04805",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | text-classification | 2022-03-02T23:29:04Z |
## Petrained Model BERT: base model (cased)
BERT base model (cased) is a pretrained model on English language using a masked language modeling (MLM) objective. It was introduced in this [paper](https://arxiv.org/abs/1810.04805) and first released in this [repository](https://github.com/google-research/bert). This model is case-sensitive: it makes a difference between english and English.
## Pretained Model Description
BERT is an auto-encoder transformer model pretrained on a large corpus of English data (English Wikipedia + Books Corpus) in a self-supervised fashion. This means the targets are computed from the inputs themselves, and humans are not needed to label the data. It was pretrained with two objectives:
- Masked language modeling (MLM)
- Next sentence prediction (NSP)
## Fine-tuned Model Description: BERT fine-tuned Cola
The pretrained model could be fine-tuned on other NLP tasks. The BERT model has been fine-tuned on a cola dataset from the GLUE BENCHAMRK, which is an academic benchmark that aims to measure the performance of ML models. Cola is one of the 11 datasets in this GLUE BENCHMARK.ย
By fine-tuning BERT on cola dataset, the model is now able to classify a given setence gramatically and semantically as acceptable or not acceptable
## How to use ?
###### Directly with a pipeline for a text-classification NLP task
```python
from transformers import pipeline
cola = pipeline('text-classification', model='Abirate/bert_fine_tuned_cola')
cola("Tunisia is a beautiful country")
[{'label': 'acceptable', 'score': 0.989352285861969}]
```
###### Breaking down all the steps (Tokenization, Modeling, Postprocessing)
```python
from transformers import AutoTokenizer, TFAutoModelForSequenceClassification
import tensorflow as tf
import numpy as np
tokenizer = AutoTokenizer.from_pretrained('Abirate/bert_fine_tuned_cola')
model = TFAutoModelForSequenceClassification.from_pretrained("Abirate/bert_fine_tuned_cola")
text = "Tunisia is a beautiful country."
encoded_input = tokenizer(text, return_tensors='tf')
#The logits
output = model(encoded_input)
#Postprocessing
probas_output = tf.math.softmax(tf.squeeze(output['logits']), axis = -1)
class_preds = np.argmax(probas_output, axis = -1)
#Predicting the class acceptable or not acceptable
model.config.id2label[class_preds]
#Result
'acceptable'
``` |
abhibisht89/spanbert-large-cased-finetuned-ade_corpus_v2 | abhibisht89 | 2021-11-21T15:23:59Z | 79 | 1 | transformers | [
"transformers",
"pytorch",
"bert",
"token-classification",
"spanbert",
"en",
"dataset:ade_corpus_v2",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | token-classification | 2022-03-02T23:29:05Z | ---
language: en
tags:
- spanbert
datasets:
- ade_corpus_v2
widget:
- text: "Having fever after taking paracetamol."
example_title: "NER"
- text: "Birth defects associated with thalidomide."
example_title: "NER"
- text: "Deafness and kidney failure associated with gentamicin (an antibiotic)."
example_title: "NER"
- text: "Bleeding of the intestine associated with aspirin therapy."
example_title: "NER"
---
spanbert-large-cased fine-tuned for <b>"Adverse drug reaction"</b> and <b>"Drug"</b> span Extraction.
<b>Details of spanbert-large-cased:</b>
https://huggingface.co/SpanBERT/spanbert-large-cased
<b>Details of the downstream task (Adverse drug reaction and Drug Extraction) - Dataset</b>
https://huggingface.co/datasets/ade_corpus_v2 |
huggingtweets/prathkum | huggingtweets | 2021-11-21T09:58:13Z | 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: https://www.huggingtweets.com/prathkum/1637488688526/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: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">๐ค AI BOT ๐ค</div>
<div style="text-align: center; font-size: 16px; font-weight: 800">Pratham</div>
<div style="text-align: center; font-size: 14px;">@prathkum</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.
| Data | Pratham |
| --- | --- |
| Tweets downloaded | 3246 |
| Retweets | 455 |
| Short tweets | 318 |
| Tweets kept | 2473 |
[Explore the data](https://wandb.ai/wandb/huggingtweets/runs/2lnm0sab/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 @prathkum's tweets.
Hyperparameters and metrics are recorded in the [W&B training run](https://wandb.ai/wandb/huggingtweets/runs/2w7zt05t) for full transparency and reproducibility.
At the end of training, [the final model](https://wandb.ai/wandb/huggingtweets/runs/2w7zt05t/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/prathkum')
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)
|
emeraldgoose/bert-base-v1-sports | emeraldgoose | 2021-11-21T05:45:05Z | 13 | 0 | transformers | [
"transformers",
"pytorch",
"bert",
"fill-mask",
"ko",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | fill-mask | 2022-03-02T23:29:05Z | ---
language: ko
mask_token: "[MASK]"
widget:
- text: ์ฐ์
์์ ๊ฑฐ ๊ฒฝ๊ธฐ๋ ์๋์ ์ผ๋ก ์๋ก์ด [MASK] 1990๋
๋์ ํ์ฑํ ๋์๋ค.
---
## Data-annotation-nlp-10 (BoostCamp AI)
์ํคํผ๋์(์คํฌ์ธ ) dataset ๊ตฌ์ถ์ ์งํํ๋ฉด์ ์ป์ ๋ฌธ์ฅ์ ํตํด bert ์ฌ์ ํ์ต์ ์งํ
## How to use
```python
from transformers import AutoTokenizer, BertForMaskedLM
model = BertForMaskedLM.from_pretrained("emeraldgoose/bert-base-v1-sports")
tokenizer = AutoTokenizer.from_pretrained("emeraldgoose/bert-base-v1-sports")
text = "์ฐ์
์์ ๊ฑฐ ๊ฒฝ๊ธฐ๋ ์๋์ ์ผ๋ก ์๋ก์ด [MASK] 1990๋
๋์ ํ์ฑํ ๋์๋ค."
inputs = tokenizer.encode(text, return_tensors='pt')
model.eval()
outputs = model(inputs)['logits']
predict = outputs.argmax(-1)[0]
print(tokenizer.decode(predict))
``` |
Leisa/marian-finetuned-kde4-en-to-fr | Leisa | 2021-11-21T05:25:45Z | 4 | 0 | transformers | [
"transformers",
"pytorch",
"marian",
"text2text-generation",
"translation",
"generated_from_trainer",
"dataset:kde4",
"license:apache-2.0",
"model-index",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | translation | 2022-03-02T23:29:04Z | ---
license: apache-2.0
tags:
- translation
- generated_from_trainer
datasets:
- kde4
metrics:
- bleu
model-index:
- name: marian-finetuned-kde4-en-to-fr
results:
- task:
name: Sequence-to-sequence Language Modeling
type: text2text-generation
dataset:
name: kde4
type: kde4
args: en-fr
metrics:
- name: Bleu
type: bleu
value: 52.94538305859332
---
<!-- 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. -->
# marian-finetuned-kde4-en-to-fr
This model is a fine-tuned version of [Helsinki-NLP/opus-mt-en-fr](https://huggingface.co/Helsinki-NLP/opus-mt-en-fr) on the kde4 dataset.
It achieves the following results on the evaluation set:
- Loss: 0.8558
- Bleu: 52.9454
## 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: 64
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 3
- mixed_precision_training: Native AMP
### Training results
### Framework versions
- Transformers 4.12.5
- Pytorch 1.10.0
- Datasets 1.15.1
- Tokenizers 0.10.3
|
Xibanya/AestheticCities | Xibanya | 2021-11-21T03:05:15Z | 0 | 0 | null | [
"region:us"
] | null | 2022-03-02T23:29:05Z | A VQGAN-compatible model trained on screenshots of cityscapes from 90s anime. To use, direct vqgan to the model as you would vqgan_imagenet_f16_1024, faceshq, etc. |
Ulto/avengers2 | Ulto | 2021-11-21T01:13:26Z | 5 | 0 | 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
datasets:
- null
model-index:
- name: avengers2
results:
- task:
name: Causal Language Modeling
type: text-generation
---
<!-- 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. -->
# avengers2
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: 4.0131
## 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 |
|:-------------:|:-----:|:----:|:---------------:|
| No log | 1.0 | 56 | 3.9588 |
| No log | 2.0 | 112 | 3.9996 |
| No log | 3.0 | 168 | 4.0131 |
### Framework versions
- Transformers 4.10.0
- Pytorch 1.9.0
- Datasets 1.2.1
- Tokenizers 0.10.1
|
arvalinno/distilbert-base-uncased-finetuned-squad | arvalinno | 2021-11-20T17:31:23Z | 6 | 0 | transformers | [
"transformers",
"pytorch",
"tensorboard",
"distilbert",
"question-answering",
"generated_from_trainer",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | question-answering | 2022-03-02T23:29:05Z | ---
license: apache-2.0
tags:
- generated_from_trainer
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 an unknown dataset.
It achieves the following results on the evaluation set:
- Loss: 1.4232
## 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 |
|:-------------:|:-----:|:-----:|:---------------:|
| 1.7604 | 1.0 | 6366 | 1.5329 |
| 1.4784 | 2.0 | 12732 | 1.3930 |
| 1.3082 | 3.0 | 19098 | 1.4232 |
### Framework versions
- Transformers 4.12.5
- Pytorch 1.10.0+cu111
- Datasets 1.15.1
- Tokenizers 0.10.3
|
Subsets and Splits