vesteinn's picture
Librarian Bot: Add base_model information to model (#2)
503900d
metadata
license: gpl-3.0
tags:
  - generated_from_trainer
datasets:
  - mim_gold_ner
metrics:
  - precision
  - recall
  - f1
  - accuracy
base_model: vesteinn/IceBERT
model-index:
  - name: IceBERT-finetuned-ner
    results:
      - task:
          type: token-classification
          name: Token Classification
        dataset:
          name: mim_gold_ner
          type: mim_gold_ner
          args: mim-gold-ner
        metrics:
          - type: precision
            value: 0.8870349771350884
            name: Precision
          - type: recall
            value: 0.8575696021029992
            name: Recall
          - type: f1
            value: 0.8720534629404617
            name: F1
          - type: accuracy
            value: 0.9848236357672584
            name: Accuracy

IceBERT-finetuned-ner

This model is a fine-tuned version of vesteinn/IceBERT on the mim_gold_ner dataset. It achieves the following results on the evaluation set:

  • Loss: 0.0815
  • Precision: 0.8870
  • Recall: 0.8576
  • F1: 0.8721
  • Accuracy: 0.9848

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
0.0536 1.0 2904 0.0749 0.8749 0.8426 0.8585 0.9831
0.0269 2.0 5808 0.0754 0.8734 0.8471 0.8600 0.9840
0.0173 3.0 8712 0.0815 0.8870 0.8576 0.8721 0.9848

Framework versions

  • Transformers 4.11.0
  • Pytorch 1.9.0+cu102
  • Datasets 1.12.1
  • Tokenizers 0.10.3