t5-small-finetuned-wikihow_3epoch_b8_lr3e-5
This model is a fine-tuned version of t5-small on the wikihow dataset. It achieves the following results on the evaluation set:
- Loss: 2.4836
- Rouge1: 25.9411
- Rouge2: 9.226
- Rougel: 21.9087
- Rougelsum: 25.2863
- Gen Len: 18.4076
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-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 |
---|---|---|---|---|---|---|---|---|
2.912 | 0.25 | 5000 | 2.6285 | 23.6659 | 7.8535 | 19.9837 | 22.9884 | 18.3867 |
2.8115 | 0.51 | 10000 | 2.5820 | 24.7979 | 8.4888 | 20.8719 | 24.1321 | 18.3292 |
2.767 | 0.76 | 15000 | 2.5555 | 25.0857 | 8.6437 | 21.149 | 24.4256 | 18.2981 |
2.742 | 1.02 | 20000 | 2.5330 | 25.3431 | 8.8393 | 21.425 | 24.7032 | 18.3749 |
2.7092 | 1.27 | 25000 | 2.5203 | 25.5338 | 8.9281 | 21.5378 | 24.9045 | 18.3399 |
2.6989 | 1.53 | 30000 | 2.5065 | 25.4792 | 8.9745 | 21.4941 | 24.8458 | 18.4565 |
2.6894 | 1.78 | 35000 | 2.5018 | 25.6815 | 9.1218 | 21.6958 | 25.0557 | 18.406 |
2.6897 | 2.03 | 40000 | 2.4944 | 25.8241 | 9.2127 | 21.8205 | 25.1801 | 18.4228 |
2.6664 | 2.29 | 45000 | 2.4891 | 25.8241 | 9.1662 | 21.7807 | 25.1615 | 18.4258 |
2.6677 | 2.54 | 50000 | 2.4855 | 25.7435 | 9.145 | 21.765 | 25.0858 | 18.4329 |
2.6631 | 2.8 | 55000 | 2.4836 | 25.9411 | 9.226 | 21.9087 | 25.2863 | 18.4076 |
Framework versions
- Transformers 4.18.0
- Pytorch 1.10.0+cu111
- Datasets 2.0.0
- Tokenizers 0.11.6
- Downloads last month
- 6
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.