News for hackers. Stuff that matters.

A comment on QwQ-32B: Embracing the Power of Reinforcement Learning. Parent comment

Re:QwQ-32B: Embracing the Power of Reinforcement Learning by AustinDev on Wednesday March 05, 2025 @07:42PM (#43274751)
I tried the 'Strawberry' question which generated nearly 70k words of CoT.
[ Parent | on HN ]
Re:QwQ-32B: Embracing the Power of Reinforcement Learning by nicman23 on Thursday March 06, 2025 @01:24AM (#43277013)
lol did it at least get it right?
[ Parent | on HN ]
Re:QwQ-32B: Embracing the Power of Reinforcement Learning by nkozyra on Thursday March 06, 2025 @09:34AM (#43280676)
It's a hard problem, that's a lot to ask.
[ Parent | on HN ]
Re:QwQ-32B: Embracing the Power of Reinforcement Learning by moffkalast on Thursday March 06, 2025 @02:55AM (#43277606)
I think you guys might be using too low of a temperature, it never goes beyond like 1k thinking tokens for me.
[ Parent | on HN ]