metadata
license: mit
tags:
- generated_from_trainer
model-index:
- name: codeparrot-ds-sample-gpt-small-10epoch
results: []
codeparrot-ds-sample-gpt-small-10epoch
This model is a fine-tuned version of gpt2 on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 2.0943
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: 0.0005
- train_batch_size: 32
- eval_batch_size: 32
- seed: 42
- gradient_accumulation_steps: 8
- total_train_batch_size: 256
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 1000
- num_epochs: 10
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 4.29 | 0.94 | 1000 | 2.8452 |
| 2.3155 | 1.88 | 2000 | 2.3659 |
| 1.8817 | 2.82 | 3000 | 2.2085 |
| 1.6245 | 3.77 | 4000 | 2.1260 |
| 1.4314 | 4.71 | 5000 | 2.0705 |
| 1.2698 | 5.65 | 6000 | 2.0603 |
| 1.1281 | 6.59 | 7000 | 2.0599 |
| 1.0108 | 7.53 | 8000 | 2.0769 |
| 0.9167 | 8.47 | 9000 | 2.0870 |
| 0.8551 | 9.42 | 10000 | 2.0943 |
Framework versions
- Transformers 4.18.0
- Pytorch 1.10.0+cu111
- Datasets 2.0.0
- Tokenizers 0.11.6