small-e-czech-finetuned-ner-wikiann

This model is a fine-tuned version of Seznam/small-e-czech on the wikiann dataset. It achieves the following results on the evaluation set:

  • Loss: 0.2547
  • Precision: 0.8713
  • Recall: 0.8970
  • F1: 0.8840
  • Accuracy: 0.9557

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: 20

Training results

Training Loss Epoch Step Validation Loss Precision Recall F1 Accuracy
0.2924 1.0 2500 0.2449 0.7686 0.8088 0.7882 0.9320
0.2042 2.0 5000 0.2137 0.8050 0.8398 0.8220 0.9400
0.1699 3.0 7500 0.1912 0.8236 0.8593 0.8411 0.9466
0.1419 4.0 10000 0.1931 0.8349 0.8671 0.8507 0.9488
0.1316 5.0 12500 0.1892 0.8470 0.8776 0.8620 0.9519
0.1042 6.0 15000 0.2058 0.8433 0.8811 0.8618 0.9508
0.0884 7.0 17500 0.2020 0.8602 0.8849 0.8724 0.9531
0.0902 8.0 20000 0.2118 0.8551 0.8837 0.8692 0.9528
0.0669 9.0 22500 0.2171 0.8634 0.8906 0.8768 0.9550
0.0529 10.0 25000 0.2228 0.8638 0.8912 0.8773 0.9545
0.0613 11.0 27500 0.2293 0.8626 0.8898 0.8760 0.9544
0.0549 12.0 30000 0.2276 0.8694 0.8958 0.8824 0.9554
0.0516 13.0 32500 0.2384 0.8717 0.8940 0.8827 0.9552
0.0412 14.0 35000 0.2443 0.8701 0.8931 0.8815 0.9554
0.0345 15.0 37500 0.2464 0.8723 0.8958 0.8839 0.9557
0.0412 16.0 40000 0.2477 0.8705 0.8948 0.8825 0.9552
0.0363 17.0 42500 0.2525 0.8742 0.8973 0.8856 0.9559
0.0341 18.0 45000 0.2529 0.8727 0.8962 0.8843 0.9561
0.0194 19.0 47500 0.2533 0.8699 0.8966 0.8830 0.9557
0.0247 20.0 50000 0.2547 0.8713 0.8970 0.8840 0.9557

Framework versions

  • Transformers 4.17.0
  • Pytorch 1.10.0+cu111
  • Datasets 1.18.4
  • Tokenizers 0.11.6
Downloads last month
11,191
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Dataset used to train richielo/small-e-czech-finetuned-ner-wikiann

Evaluation results