t5-small-mathT5-finetune_qatoexp
This model is a fine-tuned version of t5-small on the math_qa dataset. It achieves the following results on the evaluation set:
- Loss: 1.8677
- Rouge1: 21.9174
- Rouge2: 8.4401
- Rougel: 19.1645
- Rougelsum: 19.8239
- Gen Len: 18.9765
Model description
More information needed
Intended uses & limitations
More information needed
Training and evaluation data
We have trained T5-small on MathQA dataset for sequence to sequence generation of explanations from given math problem.
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 10
- eval_batch_size: 10
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 10
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Rouge1 | Rouge2 | Rougel | Rougelsum | Gen Len |
|---|---|---|---|---|---|---|---|---|
| 2.4496 | 1.0 | 2984 | 2.2096 | 19.6477 | 6.508 | 16.9295 | 17.5212 | 18.9064 |
| 2.2893 | 2.0 | 5968 | 2.0837 | 20.4879 | 7.2528 | 17.7778 | 18.4085 | 18.968 |
| 2.1869 | 3.0 | 8952 | 2.0125 | 20.8462 | 7.6105 | 18.1516 | 18.8343 | 18.9837 |
| 2.1456 | 4.0 | 11936 | 1.9633 | 20.7623 | 7.7113 | 18.1274 | 18.783 | 18.9886 |
| 2.1171 | 5.0 | 14920 | 1.9321 | 21.0648 | 7.8897 | 18.4162 | 19.0551 | 18.9844 |
| 2.0854 | 6.0 | 17904 | 1.9061 | 21.4445 | 8.0883 | 18.8038 | 19.4176 | 18.9812 |
| 2.0592 | 7.0 | 20888 | 1.8902 | 21.5714 | 8.2751 | 18.8864 | 19.537 | 18.9772 |
| 2.0609 | 8.0 | 23872 | 1.8770 | 21.7737 | 8.3297 | 19.022 | 19.6897 | 18.9763 |
| 2.0285 | 9.0 | 26856 | 1.8701 | 21.964 | 8.4358 | 19.1701 | 19.845 | 18.9747 |
| 2.0165 | 10.0 | 29840 | 1.8677 | 21.9174 | 8.4401 | 19.1645 | 19.8239 | 18.9765 |
Framework versions
- Transformers 4.22.2
- Pytorch 1.12.1+cu113
- Datasets 2.5.2
- Tokenizers 0.12.1
- Downloads last month
- 7
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.