fix: add GLM-4.6 reasoning support for OpenAI Compatible provider #9013
Conversation
- Added logic to detect GLM-4.6 models and enable thinking parameter when reasoning is enabled - The thinking parameter is now properly added for both streaming and non-streaming modes - Added comprehensive tests to verify GLM-4.6 reasoning functionality - This fixes the issue where GLM-4.6 reasoning was not working with OpenAI Compatible provider Fixes #9012
Added comment explaining why GLM-4.6 uses thinking parameter instead of reasoning_effort for better maintainability
dosubot
bot
added
the
size:L
|
Review completed. Found 1 issue that should be addressed before merging.
Mention @roomote in a comment to trigger your PR Fixer agent and make changes to this pull request. |
hannesrudolph
added
the
Issue/PR - Triage
![roomote[bot]](https://avatars.githubusercontent.com/in/1546624?s=60&v=4){kind=small}
| messages: convertedMessages, | ||
| stream: true as const, | ||
| ...(isGrokXAI ? {} : { stream_options: { include_usage: true } }), | ||
| ...(reasoning && reasoning), |
There was a problem hiding this comment.
Potential parameter conflict: The thinking parameter (added for GLM-4.6 on lines 178-180) could coexist with reasoning_effort from the spread reasoning object on this line. If a GLM-4.6 model has both supportsReasoningBinary: true and supportsReasoningEffort: true configured, both parameters would be sent to the API. Since the PR description states GLM-4.6 uses thinking instead of reasoning_effort, consider excluding the reasoning object when adding the thinking parameter to prevent both from being present simultaneously.
Fix it with Roo Code or mention @roomote and request a fix.
This PR attempts to address Issue #9012. Feedback and guidance are welcome.
Problem
GLM-4.6 model reasoning was not working correctly in v3.30.0:
Solution
Added GLM-4.6-specific reasoning support by detecting when:
enableReasoningEffort: true)supportsReasoningBinary: true)When these conditions are met, the
thinking: { type: "enabled" }parameter is added to the API request, which is the vendor-specific format required by Z AI's GLM models.Changes
thinkingparameter for both streaming and non-streaming modesTesting
Fixes #9012
Important
Adds GLM-4.6 reasoning support to OpenAI Compatible provider by introducing
thinkingparameter inOpenAiHandler.OpenAiHandlerby checking forglm-4.6in model ID,enableReasoningEffort: true, andsupportsReasoningBinary: true.thinking: { type: "enabled" }parameter to API requests when conditions are met.openai.spec.tsfor GLM-4.6 reasoning scenarios, covering both streaming and non-streaming modes.thinkingparameter is not added for non-GLM-4.6 models even with reasoning enabled.This description was created by
for c0d0e13. You can customize this summary. It will automatically update as commits are pushed.