support load model locally

[irasin]

🚀 The feature, motivation and pitch

I have downloaded the llama3-8B model from huggingface into local dir as below

But I can't run python3 torchchat.py generate --checkpoint_dir /home/Meta-Llama-3-8B-Instruct --prompt "It was a dark and stormy night, and" since it's not a single pt file.
How should we run the pre-downloaded model in torchchat?

Alternatives

No response

Additional context

No response

RFC (Optional)

No response

[Jack-Khuu]

Can you try to generate a .pth with this: https://github.com/pytorch/torchchat/blob/1566512baa889ef044231595b593749231ab0170/.ci/scripts/convert_checkpoint.sh

[irasin]

Thanks a lot.
Follow the code, I found that only ".bin" files are supported in torchchat, so I have to modify the code to support ".safetensors", and I can run the local model now.

[sunshinesfbay]

@irasin would you be able to submit a PR for docs/ADVANCED-USERS.md how to recognize and navigate this issue? Also any other issues/errors/unclear instructions you encountered?