distilbart-cnn-12-3-finetuned-pubmed
This model is a fine-tuned version of sshleifer/distilbart-cnn-12-3 on the pub_med_summarization_dataset dataset. It achieves the following results on the evaluation set:
- Loss: 2.1743
- Rouge1: 40.5642
- Rouge2: 16.9812
- Rougel: 25.3449
- Rougelsum: 36.46
- Gen Len: 141.95
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 |
|---|---|---|---|---|---|---|---|---|
| 2.469 | 1.0 | 4000 | 2.2956 | 38.3713 | 15.2594 | 23.6734 | 34.1634 | 141.707 |
| 2.2527 | 2.0 | 8000 | 2.1994 | 39.5939 | 16.2376 | 24.6363 | 35.5106 | 141.831 |
| 2.0669 | 3.0 | 12000 | 2.1780 | 40.078 | 16.6705 | 25.1119 | 35.9605 | 141.8475 |
| 1.9275 | 4.0 | 16000 | 2.1669 | 40.0825 | 16.6169 | 24.9702 | 36.0191 | 141.928 |
| 1.8102 | 5.0 | 20000 | 2.1743 | 40.5642 | 16.9812 | 25.3449 | 36.46 | 141.95 |
Framework versions
- Transformers 4.17.0
- Pytorch 1.10.0+cu111
- Datasets 1.18.3
- Tokenizers 0.11.6