t5-small-finetuned-pubmed
This model is a fine-tuned version of t5-small on the pub_med_summarization_dataset dataset. It achieves the following results on the evaluation set:
- Loss: 2.2635
- Rouge1: 8.8295
- Rouge2: 3.2594
- Rougel: 7.9975
- Rougelsum: 8.4483
- Gen Len: 19.0
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: 2
- eval_batch_size: 2
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 5
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Rouge1 | Rouge2 | Rougel | Rougelsum | Gen Len |
|---|---|---|---|---|---|---|---|---|
| 2.5892 | 1.0 | 4000 | 2.3616 | 10.1169 | 3.9666 | 8.8854 | 9.5836 | 19.0 |
| 2.559 | 2.0 | 8000 | 2.3045 | 9.4321 | 3.5398 | 8.424 | 8.984 | 19.0 |
| 2.5029 | 3.0 | 12000 | 2.2820 | 9.1658 | 3.3686 | 8.2222 | 8.7311 | 19.0 |
| 2.4673 | 4.0 | 16000 | 2.2692 | 8.8973 | 3.2617 | 8.0395 | 8.5046 | 19.0 |
| 2.4331 | 5.0 | 20000 | 2.2635 | 8.8295 | 3.2594 | 7.9975 | 8.4483 | 19.0 |
Framework versions
- Transformers 4.17.0
- Pytorch 1.10.0+cu111
- Datasets 1.18.3
- Tokenizers 0.11.6