Support the validation for reasoning #1288
Description
Description
Now many models provide reasoning_content in response. The validation should also be applied on the reasoning content
Why is this needed
- The validation of response during streaming will always fail if the LLM responds reasoning at the beginning.
- There could also be invalid or unwished contents in the reasoning, which should be blocked or filtered.
Implementation details
- Runner / AsyncRunner should be adapted to the reasoning content.
- LLMResponse should have addtional reasoning_content field
End result
Involved in release version.