Tool Calling

#1
by pelatho - opened

Is this model as good as the base model for tool calling?

I noticed it tends to send messages with JSON-formatted tool calls or HTML tool calls instead of actually properly sending tool calls. I.E, it'll just append the tool call to its message content.

I just tried the exact same prompt using the base llama 3.1 model and it worked right away. A bit strange :\

I didn't train it specifically for (or against) tool-calling! It should approximate mlabonne/Llama-3.1-70B-Instruct-lorablated, though. How does that one handle the same prompt? If it's significantly different, it would be worth retraining this IMO with some tool-calling samples.

Sign up or log in to comment