Skip to content

Conversation

@huiqiwa
Copy link

@huiqiwa huiqiwa commented Dec 26, 2025

Support tool calling for Qwen3-Coder Models

Start server
bash start_gaudi_vllm_server.sh -w /data/hf_models/Qwen3-Coder-480B-A35B-Instruct-FP8 -s -d fp8 -t 8 -e "--enable-auto-tool-choice --tool-call-parser qwen3_xml"

@huiqiwa
Copy link
Author

huiqiwa commented Dec 26, 2025

@yangulei @testdig
Please help review and merge. Many thanks.

Signed-off-by: huiqiwa <huiqi.wang@intel.com>

import regex as re

from vllm.entrypoints.chat_utils import random_tool_call_id
Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I am porting qwen3xml_tool_parser.py from v0.11.1rc5. This parser requires make_tool_call_id, which is missing in 1.22. Our branch uses random_tool_call_id.

The make_tool_call_id switch was originally introduced to support Kimi models (see this commit), but our version of the tool parser does not yet support Kimi.

I see three ways to handle this:
Modify the ported parser to use our existing random_tool_call_id.
Backport the full commit that introduced make_tool_call_id and update all existing parsers for consistency.
Add make_tool_call_id as a new helper in chat_utils.py while keeping the old function.

@czhu15 Which way do we recommend?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant