HF RL Explorer

Issue: AG-UI adapter does not display deep thinking (reasoning) content during streaming

Issue: AG-UI adapter does not display deep thinking (reasoning) content during streaming: a task in LegoFlow-SWE (Harbor dataset). When an LLM is running with deep thinking (reasoning) enabled and streaming output is active, the content generated during the thinking phase is silently discarded and…

The task

When an LLM is running with deep thinking (reasoning) enabled and streaming output is active, the content generated during the thinking phase is silently discarded and never reaches the client. Users cannot observe the model's reasoning steps.

Part of Lego-X/LegoFlow-SWE.