<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->
BART_reddit_advice_story
This model is a fine-tuned version of sshleifer/distilbart-xsum-6-6 on the None dataset. It achieves the following results on the evaluation set:
- Loss: 3.2552
- Rouge1: 21.9349
- Rouge2: 6.3417
- Rougel: 17.7133
- Rougelsum: 18.7199
- Gen Len: 21.092
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: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 3
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Rouge1 | Rouge2 | Rougel | Rougelsum | Gen Len |
|---|---|---|---|---|---|---|---|---|
| 3.3743 | 1.0 | 1875 | 3.2787 | 21.1275 | 5.9618 | 17.3772 | 18.317 | 20.447 |
| 3.025 | 2.0 | 3750 | 3.2466 | 21.8443 | 6.2351 | 17.6358 | 18.6259 | 21.506 |
| 2.7628 | 3.0 | 5625 | 3.2552 | 21.9349 | 6.3417 | 17.7133 | 18.7199 | 21.092 |
Framework versions
- Transformers 4.20.1
- Pytorch 1.12.0+cu113
- Datasets 2.3.2
- Tokenizers 0.12.1
