|
--- |
|
license: apache-2.0 |
|
tags: |
|
- generated_from_trainer |
|
model-index: |
|
- name: codeparrot-ds-500sample-gpt-neo-2ep |
|
results: [] |
|
--- |
|
|
|
<!-- This model card has been generated automatically according to the information the Trainer had access to. You |
|
should probably proofread and complete it, then remove this comment. --> |
|
|
|
# codeparrot-ds-500sample-gpt-neo-2ep |
|
|
|
This model is a fine-tuned version of [EleutherAI/gpt-neo-125M](https://huggingface.co/EleutherAI/gpt-neo-125M) on an unknown dataset. |
|
It achieves the following results on the evaluation set: |
|
- Loss: 1.5483 |
|
|
|
## 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: 2 |
|
- mixed_precision_training: Native AMP |
|
|
|
### Training results |
|
|
|
| Training Loss | Epoch | Step | Validation Loss | |
|
|:-------------:|:-----:|:-----:|:---------------:| |
|
| 4.5248 | 0.19 | 1000 | 2.9757 | |
|
| 2.5422 | 0.37 | 2000 | 2.4397 | |
|
| 2.1642 | 0.56 | 3000 | 2.1880 | |
|
| 1.9135 | 0.74 | 4000 | 1.9884 | |
|
| 1.7236 | 0.93 | 5000 | 1.8470 | |
|
| 1.5459 | 1.11 | 6000 | 1.7501 | |
|
| 1.4363 | 1.3 | 7000 | 1.6761 | |
|
| 1.3639 | 1.49 | 8000 | 1.6105 | |
|
| 1.3046 | 1.67 | 9000 | 1.5667 | |
|
| 1.273 | 1.86 | 10000 | 1.5483 | |
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.18.0 |
|
- Pytorch 1.10.0+cu111 |
|
- Datasets 2.0.0 |
|
- Tokenizers 0.11.6 |
|
|
