my_awesome_billsum_model
This model is a fine-tuned version of t5-small on the billsum dataset. It achieves the following results on the evaluation set:
- Loss: 2.3829
- Rouge1: 0.1966
- Rouge2: 0.0969
- Rougel: 0.1655
- Rougelsum: 0.1657
- 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: 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: 4
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Rouge1 | Rouge2 | Rougel | Rougelsum | Gen Len |
|---|---|---|---|---|---|---|---|---|
| No log | 1.0 | 248 | 2.5380 | 0.1441 | 0.0519 | 0.1188 | 0.1189 | 19.0 |
| No log | 2.0 | 496 | 2.4335 | 0.1939 | 0.0933 | 0.162 | 0.1622 | 19.0 |
| 2.8683 | 3.0 | 744 | 2.3940 | 0.1974 | 0.0974 | 0.1665 | 0.1666 | 19.0 |
| 2.8683 | 4.0 | 992 | 2.3829 | 0.1966 | 0.0969 | 0.1655 | 0.1657 | 19.0 |
Framework versions
- Transformers 4.25.1
- Pytorch 1.13.0+cu116
- Datasets 2.8.0
- Tokenizers 0.13.2