<!-- 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-small-gec-combine_data
This model is a fine-tuned version of t5-small on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.7615
- Rouge1: 73.8239
- Rouge2: 61.2831
- Rougel: 73.1307
- Rougelsum: 73.115
- Gen Len: 17.0115
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: 1e-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: 5
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Rouge1 | Rouge2 | Rougel | Rougelsum | Gen Len |
|---|---|---|---|---|---|---|---|---|
| 1.0791 | 0.45 | 500 | 0.8889 | 69.9014 | 56.7405 | 69.0756 | 69.0665 | 17.0912 |
| 0.9384 | 0.9 | 1000 | 0.8330 | 72.6531 | 59.6086 | 71.9311 | 71.9097 | 17.039 |
| 0.8906 | 1.35 | 1500 | 0.8059 | 73.2731 | 60.3578 | 72.5712 | 72.5505 | 17.0334 |
| 0.8764 | 1.81 | 2000 | 0.7895 | 73.552 | 60.7708 | 72.8445 | 72.8303 | 17.0228 |
| 0.852 | 2.26 | 2500 | 0.7821 | 73.6316 | 60.9269 | 72.9263 | 72.9124 | 17.0172 |
| 0.852 | 2.71 | 3000 | 0.7731 | 73.7115 | 61.0532 | 72.9932 | 72.9805 | 17.0144 |
| 0.8327 | 3.16 | 3500 | 0.7700 | 73.7518 | 61.1167 | 73.0386 | 73.0282 | 17.0169 |
| 0.8334 | 3.61 | 4000 | 0.7669 | 73.7976 | 61.2103 | 73.0951 | 73.0796 | 17.0142 |
| 0.8262 | 4.06 | 4500 | 0.7640 | 73.8157 | 61.2515 | 73.1183 | 73.1043 | 17.0126 |
| 0.8348 | 4.51 | 5000 | 0.7622 | 73.8202 | 61.2773 | 73.1264 | 73.1103 | 17.0111 |
| 0.8147 | 4.96 | 5500 | 0.7615 | 73.8239 | 61.2831 | 73.1307 | 73.115 | 17.0115 |
Framework versions
- Transformers 4.25.1
- Pytorch 1.12.1+cu113
- Datasets 2.7.1
- Tokenizers 0.13.2
