tool-call: Qwen3 Coder chat format support
#15162
Draft
+325
−7
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Related to #15012 ? |
Nepherpitou
reviewed
Aug 12, 2025
| case COMMON_CHAT_FORMAT_FUNCTIONARY_V3_1_LLAMA_3_1: return "Functionary v3.1 Llama 3.1"; | ||
| case COMMON_CHAT_FORMAT_HERMES_2_PRO: return "Hermes 2 Pro"; | ||
| case COMMON_CHAT_FORMAT_COMMAND_R7B: return "Command R7B"; | ||
| case COMMON_CHAT_FORMAT_QWEN3: return "Qwen3"; |
There was a problem hiding this comment.
I though it is Qwen3 Coder specific format, not all Qwen3 models?
|
Also check #15019. This seems to be the more proper solution since it basically implements https://huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct/blob/main/qwen3coder_tool_parser.py. It's abandoned for now though. |
|
I tried this with sst/opencode and the encoding of 
|
|
FYI, I made some changes to #15019 to make it work properly. No more tool calls in the output and works flawlessly in qwen-code and sst/opencode (a basic hello world program in C test). I've borrowed some things from this PR. 🙂 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
WIP! Will fix #15012 & #15046 when done.
Depends on Minja fixes: #15161
Qwen3 Coder has a weird syntax for tool calls: a Qwen2-esque
<tool_call>\n<function=${fn_name}>...</function></tool_call>, except the parameters aren't a JSON object. Instead they're a succession of XML-ish parameters:<parameter=${name}>\n${typeof value === 'string' : value : JSON.stringify(value)}\n</parameter>\nThis special treatment of strings technically means it's impossible for the model to output a call argument string that contains the substring
</parameter>(probably avoided having to train the model to do proper JSON string escapes?). This is quite unfortunate tbh. I don't think we should follow it to the letter. In this PR I'm just constraining the format to be:<parameter=${name}>\n${JSON.stringify(value)}\n</parameter>\n(that is, string arguments will have quotes and their content will need to be escaped)