Any guidance for further fine-tuning of LLaMA-Mesh?

[jackswl]

Great work!

Just wondering, if you can provide some form of guidance / fine-tuning template for further fine-tuning of LLaMA-Mesh for domain-specific meshes? (e.g., what exact chat template did you use for the SFT of your instruct model?) I am aware that the paper did mention about this, but having it spelt out explicitly will help tremendously especially in those possible pesky tokenization issues.

That would be really helpful.

Thanks!

[thuwzy]

Thank you for your interest in our work! As for the training code, we use LLaMA-Factory, which is a very easy-to-use codebase for LLM traninig.

[jackswl]

@thuwzy hi! thanks for your response. I mean, what kind of fine-tuning template did you use? Are you able to share the exact fine-tuning template? This is so I can conduct further fine-tuning by following your template.