viktoroo's picture
Update README.md
e743730
|
raw
history blame
3.12 kB
metadata
license: apache-2.0
tags:
  - generated_from_trainer
  - named-entity-recognition
  - russian
  - ner
language:
  - ru
thumbnail: Sberbank RuBERT-base fintuned on Collection3 dataset
datasets:
  - RCC-MSU/collection3
metrics:
  - precision
  - recall
  - f1
  - accuracy
model-index:
  - name: sberbank-rubert-base-collection3
    results:
      - task:
          name: Token Classification
          type: token-classification
        dataset:
          name: RCC-MSU/collection3
          type: named-entity-recognition
          config: default
          split: validation
          args: default
        metrics:
          - name: Precision
            type: precision
            value: 0.938019472809309
          - name: Recall
            type: recall
            value: 0.9594364828758805
          - name: F1
            type: f1
            value: 0.9486071085494716
          - name: Accuracy
            type: accuracy
            value: 0.9860420020488805
      - task:
          name: Token Classification
          type: token-classification
        dataset:
          name: RCC-MSU/collection3
          type: named-entity-recognition
          config: default
          split: test
          args: default
        metrics:
          - name: Precision
            type: precision
            value: 0.9419896321895829
          - name: Recall
            type: recall
            value: 0.9537615596100975
          - name: F1
            type: f1
            value: 0.947839046199702
          - name: Accuracy
            type: accuracy
            value: 0.9847255179564897

sberbank-rubert-base-collection3

This model is a fine-tuned version of sberbank-ai/ruBert-base on the collection3 dataset. It achieves the following results on the validation set:

  • Loss: 0.0772
  • Precision: 0.9380
  • Recall: 0.9594
  • F1: 0.9486
  • Accuracy: 0.9860

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: 4
  • eval_batch_size: 8
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: cosine
  • lr_scheduler_warmup_steps: 0.1
  • num_epochs: 5

Training results

Training Loss Epoch Step Validation Loss Precision Recall F1 Accuracy
0.0899 1.0 2326 0.0760 0.9040 0.9330 0.9182 0.9787
0.0522 2.0 4652 0.0680 0.9330 0.9339 0.9335 0.9821
0.0259 3.0 6978 0.0745 0.9308 0.9512 0.9409 0.9838
0.0114 4.0 9304 0.0731 0.9372 0.9573 0.9471 0.9857
0.0027 5.0 11630 0.0772 0.9380 0.9594 0.9486 0.9860

Framework versions

  • Transformers 4.26.1
  • Pytorch 1.7.0
  • Datasets 2.10.1
  • Tokenizers 0.13.2