SFT llama3

[joyjiaxi]

Hi,

I have two questions regarding your LLaMA-3 fine-tuning work:

1. Release of fine-tuned model:
   Would the fine-tuned LLaMA-3 model used in your experiments be publicly available? I am interested in reproducing some of your results.

2. Instruction prompt:
   I noticed that the instruction prompt used in the training script (train_lora_combined.py, line 150) differs from the prompt example shown in Table 10 of your paper (NAACL 2025, https://aclanthology.org/2025.naacl-long.451.pdf). Could you explain the reason for this difference? Is the script prompt the final one used for experiments, or just an example?

Thank you.