t5-small-finetuned-wikihow_3epoch_b8_lr3e-4
This model is a fine-tuned version of t5-small on the wikihow dataset. It achieves the following results on the evaluation set:
- Loss: 2.3136
- Rouge1: 27.3718
- Rouge2: 10.6235
- Rougel: 23.3396
- Rougelsum: 26.6889
- Gen Len: 18.5194
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: 8
- eval_batch_size: 8
- 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 |
---|---|---|---|---|---|---|---|---|
2.8029 | 0.25 | 5000 | 2.5368 | 25.2267 | 8.9048 | 21.2588 | 24.5804 | 18.4303 |
2.6924 | 0.51 | 10000 | 2.4725 | 25.6553 | 9.1904 | 21.7633 | 24.9807 | 18.5549 |
2.6369 | 0.76 | 15000 | 2.4332 | 26.2895 | 9.7203 | 22.3286 | 25.6009 | 18.4185 |
2.5994 | 1.02 | 20000 | 2.4051 | 26.1779 | 9.5708 | 22.3531 | 25.5357 | 18.561 |
2.521 | 1.27 | 25000 | 2.3805 | 26.7558 | 10.0411 | 22.7252 | 26.0476 | 18.304 |
2.5091 | 1.53 | 30000 | 2.3625 | 26.6439 | 10.0698 | 22.6662 | 25.9537 | 18.5437 |
2.4941 | 1.78 | 35000 | 2.3498 | 26.9322 | 10.2817 | 23.0002 | 26.2604 | 18.4953 |
2.4848 | 2.03 | 40000 | 2.3424 | 27.0381 | 10.3452 | 22.9749 | 26.3407 | 18.5749 |
2.4268 | 2.29 | 45000 | 2.3272 | 27.2386 | 10.4595 | 23.1866 | 26.5541 | 18.4954 |
2.4263 | 2.54 | 50000 | 2.3226 | 27.1489 | 10.532 | 23.1428 | 26.4657 | 18.5583 |
2.4161 | 2.8 | 55000 | 2.3136 | 27.3718 | 10.6235 | 23.3396 | 26.6889 | 18.5194 |
Framework versions
- Transformers 4.18.0
- Pytorch 1.10.0+cu111
- Datasets 2.0.0
- Tokenizers 0.11.6
- Downloads last month
- 15
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.