Frequent errors in tool call parameter format

[Kyrie666]

What happened?

│ x Shell {"command":"mkdir -p ai-lab-monitoring/{app,app/api,app/core,app/models,app/schemas,app/services,app/utils,tes… │
│ │
│ params/is_background must be boolean

Frequent occurrence of similar parameter format issues

What did you expect to happen?

Optimize parameter format validation and error handling methods

Client information

Client Information

Run qwen to enter the interactive CLI, then run the /about command.

$ qwen /about
# paste output here

Login information

No response

Anything else we need to know?

No response

[github-actions]

⚠️ Missing Required Information

Thank you for reporting this issue! To help us investigate and resolve this problem more effectively, we need some additional information:

• Qwen Code version
• operating system information
• full output of the /about command (recommended)

How to provide this information:

Please run the following command and paste the complete output:

qwen
# Then in the interactive CLI, run:
/about

The output should look like:

CLI Version             0.0.14
Git Commit              9a0cb64a
Model                   coder-model
Sandbox                 no sandbox
OS                      darwin
Auth Method             qwen-oauth

Once you provide this information, we'll be able to assist you better. Thank you! 🙏

[tanzhenxin]

@Kyrie666 Thanks for reporting this issue! Could you please attach the output of /about command so that we can better troubleshoot this problem?

[Kyrie666]

@Kyrie666 Thanks for reporting this issue! Could you please attach the output of /about command so that we can better troubleshoot this problem?

CLI Version 0.2.0 │
│ Git Commit 7a472e4 │
│ Model qwen3-coder-480b-a35b-fp8 │
│ Sandbox no sandbox
Auth Method openai
OS Platform linux │
│ OS Arch x64

[tanzhenxin]

@Kyrie666 It is hard to tell from the output but are you using self hosted inference service? If it is the case, pls refer to our External LLM User Notice for more details.

[Kyrie666]

@Kyrie666 It is hard to tell from the output but are you using self hosted inference service? If it is the case, pls refer to our External LLM User Notice for more details.

it's qwen-coder-480B that's deployed by myself。But obviously I believe this issue can be optimized. The main problem is that the model outputs strings, while tool calls require boolean values or specific formats.

[tanzhenxin]

@Kyrie666 It is hard to tell from the output but are you using self hosted inference service? If it is the case, pls refer to our External LLM User Notice for more details.

it's qwen-coder-480B that's deployed by myself。But obviously I believe this issue can be optimized. The main problem is that the model outputs strings, while tool calls require boolean values or specific formats.

While I’m not completely certain, this issue is most likely caused by inference engines like vLLM or llama.cpp, which have limited support for the tool output formats of qwen3-coder-model.
You can refer to this issue for similar reports.

[Gerrard-YNWA]

I went the same error with Qwen3-Coder-30B-A3B-Instruct on vllm too.
My fix is from the vllm tool-calling doc. I upgrade vllm to v0.12.0 and switching the --tool-call-parser to qwen3_xml works as expected.