The notebook to train the model is available here. it is the Llama 3.2 1B base model (the unsloth Version) finetuned to write Python code with the iamtarun/python_code_instructions_18k_alpaca dataset.

I wrote about the process on my blog, here.

This llama model was trained 2x faster with Unsloth and Huggingface's TRL library.