distilbart-cnn-6-6-finetuned-pubmed
This model is a fine-tuned version of sshleifer/distilbart-cnn-6-6 on the pub_med_summarization_dataset dataset. It achieves the following results on the evaluation set:
- Loss: 2.0648
- Rouge1: 39.2769
- Rouge2: 15.876
- Rougel: 24.2306
- Rougelsum: 35.267
- Gen Len: 141.8565
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 |
|---|---|---|---|---|---|---|---|---|
| 2.2215 | 1.0 | 4000 | 2.0781 | 37.2476 | 14.2852 | 22.6875 | 33.1607 | 141.97 |
| 2.0105 | 2.0 | 8000 | 2.0217 | 37.8038 | 14.7869 | 23.2025 | 33.7069 | 141.918 |
| 1.8331 | 3.0 | 12000 | 2.0243 | 39.0497 | 15.8077 | 24.2237 | 34.9371 | 141.822 |
| 1.6936 | 4.0 | 16000 | 2.0487 | 38.7059 | 15.4364 | 23.8514 | 34.7771 | 141.878 |
| 1.5817 | 5.0 | 20000 | 2.0648 | 39.2769 | 15.876 | 24.2306 | 35.267 | 141.8565 |
Framework versions
- Transformers 4.17.0
- Pytorch 1.10.0+cu111
- Datasets 1.18.3
- Tokenizers 0.11.6