g8a9's picture
update model card README.md
8e7fcb7
metadata
license: mit
tags:
  - generated_from_trainer
model-index:
  - name: gpt2-kl_1_06-hs_cn_decay
    results: []

gpt2-kl_1_06-hs_cn_decay

This model is a fine-tuned version of gpt2-medium on an unknown dataset. It achieves the following results on the evaluation set:

  • Loss: 0.5526

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: 8
  • eval_batch_size: 4
  • seed: 21
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_ratio: 0.1
  • num_epochs: 3.0

Training results

Training Loss Epoch Step Validation Loss
74.1745 0.02 10 69.5746
46.6776 0.04 20 32.9878
14.1513 0.06 30 10.6492
6.163 0.08 40 4.4128
3.2475 0.1 50 1.9870
1.8729 0.12 60 1.1410
1.6096 0.14 70 0.8221
1.354 0.16 80 0.7575
1.3447 0.18 90 0.6948
1.2679 0.2 100 0.6150
1.1468 0.22 110 0.5965
1.1623 0.24 120 0.5873
1.1934 0.26 130 0.5940
1.2443 0.28 140 0.6013
1.2871 0.3 150 0.5707
1.162 0.32 160 0.5647
1.0518 0.34 170 0.5655
1.2081 0.36 180 0.5626
1.2465 0.38 190 0.5582
1.0764 0.4 200 0.5575
1.2877 0.42 210 0.5528
1.2087 0.44 220 0.5506
1.0139 0.46 230 0.5549
1.0924 0.48 240 0.5516
1.0167 0.5 250 0.5526

Framework versions

  • Transformers 4.28.0
  • Pytorch 1.11.0+cu113
  • Datasets 2.11.0
  • Tokenizers 0.12.1