Edit model card

t5-small-finetuned-wikihow_3epoch_v2

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

  • Loss: 2.2758
  • Rouge1: 27.48
  • Rouge2: 10.7621
  • Rougel: 23.4136
  • Rougelsum: 26.7923
  • Gen Len: 18.5424

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: 0.0003
  • 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
2.8423 0.13 5000 2.5715 25.2685 8.6964 21.229 24.5773 18.4479
2.7345 0.25 10000 2.5236 24.982 8.7823 21.1609 24.3066 18.3631
2.6811 0.38 15000 2.4911 25.7585 9.3372 21.8388 25.1052 18.3997
2.6611 0.51 20000 2.4510 26.022 9.4708 22.0899 25.3236 18.5472
2.6133 0.64 25000 2.4272 26.3481 9.6769 22.4484 25.7046 18.3863
2.6083 0.76 30000 2.4108 26.4131 9.6643 22.4021 25.6958 18.5585
2.5842 0.89 35000 2.3866 26.2852 9.7505 22.4525 25.5908 18.5485
2.5554 1.02 40000 2.3816 26.3018 9.7218 22.3673 25.6515 18.4912
2.4895 1.14 45000 2.3730 26.6439 9.9665 22.6593 25.9521 18.5635
2.4781 1.27 50000 2.3541 26.8488 10.0364 22.8202 26.1598 18.4254
2.4821 1.4 55000 2.3440 26.9511 10.2079 23.0133 26.2821 18.5712
2.4593 1.53 60000 2.3370 26.945 10.3123 22.9245 26.2493 18.5978
2.4521 1.65 65000 2.3309 26.9652 10.314 22.9657 26.298 18.4837
2.4523 1.78 70000 2.3249 27.0548 10.4204 23.1286 26.379 18.4717
2.4563 1.91 75000 2.3079 27.4563 10.6452 23.3985 26.7812 18.5642
2.4229 2.03 80000 2.3115 27.0538 10.44 22.9957 26.349 18.5914
2.3694 2.16 85000 2.3017 27.332 10.6556 23.3135 26.629 18.459
2.3749 2.29 90000 2.2941 27.3294 10.5967 23.2039 26.6411 18.5179
2.3779 2.42 95000 2.2891 27.3725 10.6539 23.3455 26.707 18.5367
2.3638 2.54 100000 2.2895 27.3487 10.6738 23.2894 26.681 18.6128
2.3549 2.67 105000 2.2833 27.408 10.6903 23.3575 26.7137 18.6035
2.3652 2.8 110000 2.2788 27.561 10.8202 23.4672 26.8584 18.5565
2.3553 2.93 115000 2.2758 27.48 10.7621 23.4136 26.7923 18.5424

Framework versions

  • Transformers 4.17.0
  • Pytorch 1.10.0+cu111
  • Datasets 2.0.0
  • Tokenizers 0.11.6
Downloads last month
13

Dataset used to train Sevil/t5-small-finetuned-wikihow_3epoch_v2

Evaluation results