| base_model: meta-llama/Llama-3.2-3B-Instruct | |
| datasets: | |
| - tatsu-lab/alpaca | |
| language: en | |
| tags: | |
| - torchtune | |
| # my_cool_model | |
| This model is a finetuned version of [meta-llama/Llama-3.2-3B-Instruct](https://huggingface.co/meta-llama/Llama-3.2-3B-Instruct) on the [tatsu-lab/alpaca](https://huggingface.co/tatsu-lab/alpaca) dataset. | |
| # Model description | |
| More information needed | |
| # Training and evaluation results | |
| More information needed | |
| # Training procedure | |
| This model was trained using the [torchtune](https://github.com/pytorch/torchtune) library using the following command: | |
| ```bash | |
| ppo_full_finetune_single_device.py\ | |
| --config ./target/7B_full_ppo_low_memory_single_device.yaml\ | |
| device=cuda\ | |
| metric_logger._component_=torchtune.utils.metric_logging.WandBLogger\ | |
| metric_logger.project=torchtune_ppo\ | |
| forward_batch_size=2\ | |
| batch_size=64\ | |
| ppo_batch_size=32\ | |
| gradient_accumulation_steps=16\ | |
| compile=True\ | |
| optimizer._component_=bitsandbytes.optim.PagedAdamW\ | |
| optimizer.lr=3e-4 | |
| ``` | |
| # Framework versions | |
| - torchtune | |
| - torchao 0.5.0 | |
| - datasets 2.20.0 | |
| - sentencepiece 0.2.0 | |