mega-ar-126m-v12-python-apps-4096
This model is a fine-tuned version of pszemraj/mega-ar-126m-v12-KIx3 on the None dataset. It achieves the following results on the evaluation set:
- Loss: 1.3247
- Accuracy: 0.7266
Model description
Just a test - dataset it was tuned on is rather narrow. Note that this has 4096ctx so may be worth giving it 2k tokens or so and seeing how it completes that
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 0.0002
- train_batch_size: 4
- eval_batch_size: 4
- seed: 7427
- gradient_accumulation_steps: 32
- total_train_batch_size: 128
- optimizer: Adam with betas=(0.9,0.98) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.05
- num_epochs: 2.0
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|---|---|---|---|---|
| 1.2235 | 0.32 | 50 | 1.5113 | 0.6991 |
| 1.1655 | 0.64 | 100 | 1.4277 | 0.7109 |
| 1.1171 | 0.96 | 150 | 1.3812 | 0.7183 |
| 1.0725 | 1.28 | 200 | 1.3539 | 0.7220 |
| 1.0304 | 1.6 | 250 | 1.3356 | 0.7246 |
| 0.9842 | 1.92 | 300 | 1.3247 | 0.7266 |
Framework versions
- Transformers 4.33.3
- Pytorch 2.2.0.dev20231017+cu121
- Datasets 2.14.5
- Tokenizers 0.13.3
- Downloads last month
- 1