distilgpt2-fables-demo
Training: The model has been trained using the script provided in the following repository https://github.com/MorenoLaQuatra/transformers-tasks-templates
This model is a fine-tuned version of distilgpt2 on demelin/understanding_fables dataset. It achieves the following results on the evaluation set:
- Loss: 3.2165
Model description
The model is a demo for the fine-tuning of decoder-only models using transformers library.
Intended uses & limitations
It can be used mainly for prototyping and educational purposes.
Training and evaluation data
The demelin/understanding_fables dataset has been split into train/test/validation using an 80/10/10 random split (random_seed = 42). Google Colab has been used for model fine-tuning.
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 4
- eval_batch_size: 4
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 10
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| No log | 1.0 | 38 | 42.4563 |
| No log | 2.0 | 76 | 5.2808 |
| 28.753 | 3.0 | 114 | 3.7712 |
| 28.753 | 4.0 | 152 | 3.4577 |
| 28.753 | 5.0 | 190 | 3.3120 |
| 3.5846 | 6.0 | 228 | 3.2773 |
| 3.5846 | 7.0 | 266 | 3.2710 |
| 3.0017 | 8.0 | 304 | 3.2764 |
| 3.0017 | 9.0 | 342 | 3.2795 |
| 3.0017 | 10.0 | 380 | 3.3300 |
Framework versions
- Transformers 4.22.1
- Pytorch 1.12.1+cu113
- Datasets 2.4.0
- Tokenizers 0.12.1