mbert-finetuned-ner
This model is a fine-tuned version of bert-base-multilingual-cased on the wikiann dataset. It achieves the following results on the evaluation set:
- Loss: 0.1264
- Precision: 0.9305
- Recall: 0.9375
- F1: 0.9340
- Accuracy: 0.9700
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 | Precision | Recall | F1 | Accuracy |
---|---|---|---|---|---|---|---|
0.301 | 1.0 | 625 | 0.1756 | 0.8843 | 0.9067 | 0.8953 | 0.9500 |
0.1259 | 2.0 | 1250 | 0.1248 | 0.9285 | 0.9335 | 0.9310 | 0.9688 |
0.0895 | 3.0 | 1875 | 0.1264 | 0.9305 | 0.9375 | 0.9340 | 0.9700 |
Framework versions
- Transformers 4.19.4
- Pytorch 1.11.0+cu113
- Datasets 2.2.2
- Tokenizers 0.12.1
- Downloads last month
- 112
Inference Providers
NEW
This model is not currently available via any of the supported Inference Providers.
Dataset used to train Andrey1989/mbert-finetuned-ner
Evaluation results
- Precision on wikiannself-reported0.930
- Recall on wikiannself-reported0.938
- F1 on wikiannself-reported0.934
- Accuracy on wikiannself-reported0.970