Edit model card

t5-small-finetuned-wikihow_3epoch_b4_lr3e-5

This model is a fine-tuned version of t5-small on the wikihow dataset. It achieves the following results on the evaluation set:

  • Loss: 2.4351
  • Rouge1: 26.1071
  • Rouge2: 9.3627
  • Rougel: 22.0825
  • Rougelsum: 25.4514
  • Gen Len: 18.474

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: 3e-05
  • train_batch_size: 4
  • eval_batch_size: 4
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • num_epochs: 3
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Rouge1 Rouge2 Rougel Rougelsum Gen Len
2.9216 0.13 5000 2.6385 23.8039 7.8863 20.0109 23.0802 18.3481
2.8158 0.25 10000 2.5884 24.2567 8.2003 20.438 23.5325 18.3833
2.7743 0.38 15000 2.5623 24.8471 8.3768 20.8711 24.1114 18.2901
2.7598 0.51 20000 2.5368 25.1566 8.6721 21.1896 24.4558 18.3561
2.7192 0.64 25000 2.5220 25.3477 8.8106 21.3799 24.6742 18.3108
2.7207 0.76 30000 2.5114 25.5912 8.998 21.5508 24.9344 18.3445
2.7041 0.89 35000 2.4993 25.457 8.8644 21.4516 24.7965 18.4354
2.687 1.02 40000 2.4879 25.5886 8.9766 21.6794 24.9512 18.4035
2.6652 1.14 45000 2.4848 25.7367 9.078 21.7096 25.0924 18.4328
2.6536 1.27 50000 2.4761 25.7368 9.1609 21.729 25.0866 18.3117
2.6589 1.4 55000 2.4702 25.7738 9.1413 21.7492 25.114 18.4862
2.6384 1.53 60000 2.4620 25.7433 9.1356 21.8198 25.0896 18.489
2.6337 1.65 65000 2.4595 26.0919 9.2605 21.9447 25.4065 18.4083
2.6375 1.78 70000 2.4557 26.0912 9.3469 22.0182 25.4428 18.4133
2.6441 1.91 75000 2.4502 26.1366 9.3143 22.058 25.4673 18.4972
2.6276 2.03 80000 2.4478 25.9929 9.2464 21.9271 25.3263 18.469
2.6062 2.16 85000 2.4467 26.0465 9.3166 22.0342 25.3998 18.3777
2.6126 2.29 90000 2.4407 26.1953 9.3848 22.1148 25.5161 18.467
2.6182 2.42 95000 2.4397 26.1331 9.3626 22.1076 25.4627 18.4413
2.6041 2.54 100000 2.4375 26.1301 9.3567 22.0869 25.465 18.4929
2.5996 2.67 105000 2.4367 26.0956 9.3314 22.063 25.4242 18.5074
2.6144 2.8 110000 2.4355 26.1764 9.4157 22.1231 25.5175 18.4729
2.608 2.93 115000 2.4351 26.1071 9.3627 22.0825 25.4514 18.474

Framework versions

  • Transformers 4.18.0
  • Pytorch 1.10.0+cu111
  • Datasets 2.0.0
  • Tokenizers 0.11.6
Downloads last month
5
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 Chikashi/t5-small-finetuned-wikihow_3epoch_b4_lr3e-5

Evaluation results