There are a couple of issues in the reasoning output for models like R1 in assistant:
- Formatting for reasoning tokens while streaming is not applied. If you wait until reasoning has finished and expand, you see formatting properly.
- When expanding the reasoning section, there is no clear visual delimiter between where the reasoning ends and the final answer begins.
I expect:
- Reasoning tokens to get formatted in real time, including preserved newlines, as the output is streamed
- There to be a clear visual delimiter between where the reasoning section ends and the final answer begins
