Option for collapsing `<think>...</think>` inside some `<details>` tag in the `ChatInterface` in the default `load_chat` (to not require extra-coding for this very wide-spread micro-format)

[vadimkantorov]

It is useful, as with extremely long reasoning outputs, browser scrollbar becomes quite useless

It would also be useful to have a side panel with the messages list to allow to navigate between LLM model outputs and user inputs

[freddyaboulton]

The UI supports this but it's not done by default. You need to put the thinking block in a separate message with metadata: {'title': "<your-title>"}

See for example demo/chatbot_thoughts or the smolagents UI:

[vadimkantorov]

I think given the wide-spreadness of <think> and </think> micro-format, it would be very nice to extend the default load_chat to provide this option (collapsing contents of think contents) directly without extra coding

It might also be nice to create a CLI version of load_chat:

• https://github.com/gradio-app/gradio/issues/11334

[vadimkantorov]

I.e. I propose to extend the OpenAI-talking client in external.py/load_chat to optionally also use this technique and do this thinking message extraction as in https://github.com/gradio-app/gradio/blob/477730ef51697a355a09020b235f6cc4a6fbb9dc/demo/chatbot_thoughts/run.py#L30-L39

[abidlabs]

I think it's reasonable for us to consider supporting this by default in gr.load_chat -- I'll reopen @freddyaboulton @aliabid94