<!-- 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. -->
t5-base-finetuned-scitldr-only-abstract
This model is a fine-tuned version of t5-base on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 2.3365
- Rouge1: 34.3531
- Rouge2: 15.7554
- Rougel: 29.8918
- Rougelsum: 29.9514
- Gen Len: 18.7658
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: 3e-06
- 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: 2
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Rouge1 | Rouge2 | Rougel | Rougelsum | Gen Len |
|---|---|---|---|---|---|---|---|---|
| 2.76 | 1.0 | 996 | 2.3649 | 34.0043 | 15.5031 | 29.4997 | 29.5576 | 18.7835 |
| 2.4843 | 2.0 | 1992 | 2.3365 | 34.3531 | 15.7554 | 29.8918 | 29.9514 | 18.7658 |
Framework versions
- Transformers 4.18.0
- Pytorch 1.10.0+cu111
- Datasets 2.0.0
- Tokenizers 0.11.6
