Configurability

[tumblewader]

Now that we are starting to get open-source reasoning models we might have to make things more configurable for the user.

For example, deepseek-r1 recommends that.

• Set the temperature within the range of 0.5-0.7 (0.6 is recommended) to prevent endless repetitions or incoherent outputs.
• Avoid adding a system prompt; all instructions should be contained within the user prompt.

I think these setting could be part of the model configuration screens, moreover for open router provider i think the choice of a provider is very much needed to filter out providers which are using for example distilled versions or less context.

[BradKML]

What is the current recommendation for DeepSeek and Qwen?