QwQ-32B: Embracing the Power of Reinforcement Learning

Re:QwQ-32B: Embracing the Power of Reinforcement Learning by zamadatix on Wednesday March 05, 2025 @09:35PM (#43275735)

I always see:

  /set parameter num_ctx <value>

Explained but never the follow up:

  /save <custom-name>

So you don't have to do the parameter change every load. Is there a better way or is it kind of like setting num_ctx in that "you're just supposed to know"?

[ Parent | on HN ]

Re:QwQ-32B: Embracing the Power of Reinforcement Learning by sReinwald on Thursday March 06, 2025 @03:17AM (#43277757)

You can also set

    OLLAMA_CONTEXT_LENGTH=<tokens>

as an environment variable to change ollama's default context length.

[ Parent | on HN ]

Re:QwQ-32B: Embracing the Power of Reinforcement Learning by Tepix on Thursday March 06, 2025 @08:27AM (#43279944)

I think that will not work if you use the OpenAI compatible API endpoint.

[ Parent | on HN ]

Re:QwQ-32B: Embracing the Power of Reinforcement Learning by svachalek on Thursday March 06, 2025 @10:14AM (#43281153)

I tried this with ollama run, and it had no effect at all.

[ Parent | on HN ]

Re:QwQ-32B: Embracing the Power of Reinforcement Learning by underlines on Thursday March 06, 2025 @05:04PM (#43285513)

that env parameter is brand new, did you update ollama?

[ Parent | on HN ]