<!-- 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_gaming
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: 3.7373
- Rouge1: 18.1202
- Rouge2: 4.6045
- Rougel: 15.1273
- Rougelsum: 15.7601
- Gen Len: 18.208
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 |
|---|---|---|---|---|---|---|---|---|
| 3.864 | 1.0 | 1875 | 3.7752 | 17.3754 | 4.51 | 14.6763 | 15.22 | 16.944 |
| 3.4755 | 2.0 | 3750 | 3.7265 | 17.8066 | 4.4188 | 14.9432 | 15.5396 | 18.104 |
| 3.2629 | 3.0 | 5625 | 3.7373 | 18.1202 | 4.6045 | 15.1273 | 15.7601 | 18.208 |
Framework versions
- Transformers 4.20.1
- Pytorch 1.12.0+cu113
- Datasets 2.3.2
- Tokenizers 0.12.1
