Bug fix for the tool use chat template.

#2
MLX Community org
edited Mar 3

The chat template provides an example for how tool use should be formatted. In the current version of the tokenizer_config file, these tools are mistakenly wrapped in double curly braces. This confuses smaller models (tested on the 14B and 7B versions).

The proposed changed also matches the original chat template in the official Qwen2.5 repo, see: https://huggingface.co/Qwen/Qwen2.5-32B-Instruct/blob/main/tokenizer_config.json

Ready to merge
This branch is ready to get merged automatically.

Sign up or log in to comment