Regis David Souza Mesquita
a4f110f80a
Fixes the logic to consider only the last request. It is not really the thought time but the whole request time, but it is nice anyway :) feel free to change the wording or just close this PR. just something that I did to use locally and thought people could use as well.
Better-R1
An open webui function for better local R1 experience
This is a simple open webui function for R1 models, it can do the following:
- Replace the simple
<think>tags with<details>&<summary>tags, which makes R1's thoughts collapsible, like this:
Thoughts💬
Emmm.. LASAGNA...- Remove R1's old thoughts in multi-turn conversation, according to deepseeks API docs you should always remove R1's previous thoughts in a multi-turn conversation.
Note: This function is only designed for those who run R1 (-distilled) models locally, such as using Ollama. It does not work with the DeepSeek API.
Code: https://github.com/AaronFeng753/Better-R1/blob/main/BetterR1.txt
Check out my other project: https://github.com/AaronFeng753/Waifu2x-Extension-GUI
DEMO:
