parameter.enable_thinking must be set to false

[lai-vson]

Git provider (optional)

None

System Info (optional)

Qwen3 model

Issues details

Is there a way to set parameter.enable_thinking to false currently? We are using Qwen3 model which came out this error.

[qodo-merge-for-open-source]

💡 Auto-generated documentation-based answer:

Based on the documentation, there isn't a specific parameter.enable_thinking configuration option mentioned. However, for reasoning models like Qwen3 that don't support chat-style inputs or temperature settings, you should set:

[config]
custom_reasoning_model = true

This configuration bypasses chat templates and temperature controls for reasoning models. Add this to your configuration file (.pr_agent.toml or configuration.toml).

If you're still experiencing issues with the Qwen3 model, you may also want to check:

1. Your model configuration in the [config] section
2. Ensure you have the correct API keys and environment variables set for your model provider
3. Verify that your model name is correctly specified in the configuration

For more details on model configuration, see the "Changing a Model" documentation section.

Relevant Sources:

• https://qodo-merge-docs.qodo.ai//usage-guide/changing_a_model#custom-models

[lai-vson]

Still encountering this issue after setting custom reasoning model to true via env var.

[mrT23]

i don't understand which parameter you are talking about.

Share dramatically more details, the full error message, environment details, and more.

The error you are talking about is due to your specific unique setup, and not a pr-agent error

[lai-vson]

Currently i had do a local docker setup for gitlab webhook. The model i had been using is Qwen3. I had dig deeper into the error message. There is a thinking mode and non-thinking mode in Qwen3 model and causing this error. I'm not sure if there is any relevant config i can do in pr agent for now. Below are the documentations that I found. However, i had now revert back to older model which does not have this thinking mode.

https://qwen.readthedocs.io/en/latest/deployment/vllm.html

[mrT23]

You are welcome to open a PR and add better support. Some models need additional params.
https://github.com/qodo-ai/pr-agent/blob/main/pr_agent/algo/ai_handlers/litellm_ai_handler.py

Also look here and see if it is supported:
BerriAI/litellm#10938