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: 2.7576
- Rouge1: 0.1327
- Rouge2: 0.0444
- Rougel: 0.1111
- Rougelsum: 0.1111
- 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: 16
- eval_batch_size: 16
- 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 | 62 | 3.0485 | 0.1269 | 0.0387 | 0.1064 | 0.1065 | 19.0 |
| No log | 2.0 | 124 | 2.8371 | 0.1322 | 0.0468 | 0.1114 | 0.1114 | 19.0 |
| No log | 3.0 | 186 | 2.7747 | 0.1335 | 0.0464 | 0.1124 | 0.1123 | 19.0 |
| No log | 4.0 | 248 | 2.7576 | 0.1327 | 0.0444 | 0.1111 | 0.1111 | 19.0 |
Framework versions
- Transformers 4.25.1
- Pytorch 1.13.1+cu117
- Datasets 2.8.0
- Tokenizers 0.13.2