distilbert-srb-ner

This model was trained from scratch on the wikiann dataset. It achieves the following results on the evaluation set:

  • Loss: 0.2972
  • Precision: 0.8871
  • Recall: 0.9100
  • F1: 0.8984
  • Accuracy: 0.9577

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: 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
0.3818 1.0 625 0.2175 0.8175 0.8370 0.8272 0.9306
0.198 2.0 1250 0.1766 0.8551 0.8732 0.8640 0.9458
0.1423 3.0 1875 0.1702 0.8597 0.8763 0.8679 0.9473
0.079 4.0 2500 0.1774 0.8674 0.8875 0.8773 0.9515
0.0531 5.0 3125 0.2011 0.8688 0.8965 0.8825 0.9522
0.0429 6.0 3750 0.2082 0.8769 0.8970 0.8868 0.9538
0.032 7.0 4375 0.2268 0.8764 0.8916 0.8839 0.9528
0.0204 8.0 5000 0.2423 0.8726 0.8959 0.8841 0.9529
0.0148 9.0 5625 0.2522 0.8774 0.8991 0.8881 0.9538
0.0125 10.0 6250 0.2544 0.8823 0.9024 0.8922 0.9559
0.0108 11.0 6875 0.2592 0.8780 0.9041 0.8909 0.9553
0.007 12.0 7500 0.2672 0.8877 0.9056 0.8965 0.9571
0.0048 13.0 8125 0.2714 0.8879 0.9089 0.8982 0.9583
0.0049 14.0 8750 0.2872 0.8873 0.9068 0.8970 0.9573
0.0034 15.0 9375 0.2915 0.8883 0.9114 0.8997 0.9577
0.0027 16.0 10000 0.2890 0.8865 0.9103 0.8983 0.9581
0.0028 17.0 10625 0.2885 0.8877 0.9085 0.8980 0.9576
0.0014 18.0 11250 0.2928 0.8860 0.9073 0.8965 0.9577
0.0013 19.0 11875 0.2963 0.8856 0.9099 0.8976 0.9576
0.001 20.0 12500 0.2972 0.8871 0.9100 0.8984 0.9577

Framework versions

  • Transformers 4.9.2
  • Pytorch 1.9.0
  • Datasets 1.11.0
  • Tokenizers 0.10.1
Downloads last month
19
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 Aleksandar/distilbert-srb-ner