Edit model card

electra-srb-ner

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

  • Loss: 0.3406
  • Precision: 0.8934
  • Recall: 0.9087
  • F1: 0.9010
  • Accuracy: 0.9568

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
0.3686 1.0 625 0.2108 0.8326 0.8494 0.8409 0.9335
0.1886 2.0 1250 0.1784 0.8737 0.8713 0.8725 0.9456
0.1323 3.0 1875 0.1805 0.8654 0.8870 0.8760 0.9468
0.0675 4.0 2500 0.2018 0.8736 0.8880 0.8807 0.9502
0.0425 5.0 3125 0.2162 0.8818 0.8945 0.8881 0.9512
0.0343 6.0 3750 0.2492 0.8790 0.8928 0.8859 0.9513
0.0253 7.0 4375 0.2562 0.8821 0.9006 0.8912 0.9525
0.0142 8.0 5000 0.2788 0.8807 0.9013 0.8909 0.9524
0.0114 9.0 5625 0.2793 0.8861 0.9002 0.8931 0.9534
0.0095 10.0 6250 0.2967 0.8887 0.9034 0.8960 0.9550
0.008 11.0 6875 0.2993 0.8899 0.9067 0.8982 0.9556
0.0048 12.0 7500 0.3215 0.8887 0.9038 0.8962 0.9545
0.0034 13.0 8125 0.3242 0.8897 0.9068 0.8982 0.9554
0.003 14.0 8750 0.3311 0.8884 0.9085 0.8983 0.9559
0.0025 15.0 9375 0.3383 0.8943 0.9062 0.9002 0.9562
0.0011 16.0 10000 0.3346 0.8941 0.9112 0.9026 0.9574
0.0015 17.0 10625 0.3362 0.8944 0.9081 0.9012 0.9567
0.001 18.0 11250 0.3464 0.8877 0.9100 0.8987 0.9559
0.0012 19.0 11875 0.3415 0.8944 0.9089 0.9016 0.9568
0.0005 20.0 12500 0.3406 0.8934 0.9087 0.9010 0.9568

Framework versions

  • Transformers 4.9.2
  • Pytorch 1.9.0
  • Datasets 1.11.0
  • Tokenizers 0.10.1
Downloads last month
3
Safetensors
Model size
109M params
Tensor type
I64
·
F32
·
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/electra-srb-ner