search

meta-llama / llama-stack

Composable building blocks to build Llama Apps
MIT License
4.43k stars 558 forks source link

LLM hallucinating while tool calling #124

Open cheesecake100201 opened 1 month ago

cheesecake100201 commented 1 month ago

image The tool call shows something called as brary search which neither I have passed nor exists. I have only passed inbuilt tools which are brave_Search, wolfram alpha and code interpreter. I am using the llama-3.1-405b-instruct model for testing through the fireworks inference API. I have added debug statements to check the tools being passed before and after the chat completion API is called and it shows that the tools being passed are accurate and as in the jupyter notebook.

TurboMa commented 20 hours ago

Do you use the native tool calling capability from the model itself?