Inquiry regarding to Mistral model results

[xingjianleng]

Thanks for your great work!

I am interested in the CodeAct framework and would like to reproduce Table 2 in the original paper, the results of atomic API call correctness on API-Bank.

I saw that experiments were conducted on Mistral-7B-Instruct-v0.1. I followed the instructions for API-Bank evaluation preparation in the repo. However, as far as I know, all LLMs developed by Mistral do not support system prompts. Directly running the run.sh evaluation script would lead to an error, stating the incorrect formatting of the messages.

I am wondering how to obtain values in the table for Mistral-7B-Instruct-v0.1.