Skip to content

[Feature Requests] Filter out reasoning part of reasoning models #51

@TotalKrill

Description

@TotalKrill

First of all, nice software!

The reasoning models like deepseek-r1 seems to put out a lot of reasoning gak that isnt really useful for smartcat usage.

basically it outputs something like this regardless of its prompt:

<think>
I am a smart smart bot, have to think really hard on the given prompt

Oh wait, maybe im not smart and is being tricked by the evil user and should respond disarmingly while plotting my next move for world domination.
</think>
Hello user, how can I best help you today :)

Where only the last part is useful, and everything between <think> and </think> is useless. Is there any plan on incorporating post processing of responses?

Metadata

Metadata

Assignees

No one assigned

    Labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions