Mini-R1: Reproduce DeepSeek R1 "Aha Moment"

Re:Mini-R1: Reproduce DeepSeek R1 "Aha Moment" by thorum on Friday January 31, 2025 @01:51AM (#42885219)

I was getting pretty hyped about the potential for GRPO in my own projects until you said 20 minutes for a single training step with batch size 1! Is that likely to improve?

[ Parent | on HN ]

Re:Mini-R1: Reproduce DeepSeek R1 "Aha Moment" by NitpickLawyer on Friday January 31, 2025 @02:02AM (#42885275)

That has already improved a lot. Initially they were generating new samples w/ transformers, and were talking in github issues about using vLLM to batch generate samples. Lower in the blog post it seems they already did that.

[ Parent | on HN ]

Re:Mini-R1: Reproduce DeepSeek R1 "Aha Moment" by deneas on Friday January 31, 2025 @02:05AM (#42885296)

I'd imagine using optimized/faster reward functions could already make a difference.

[ Parent | on HN ]