policy-gradient/GRPO-Zero

Implementing DeepSeek R1's GRPO algorithm from scratch

192 best points · 3 comments · 1 Hacker News post · 1,912 GitHub stars · View on GitHub

Hacker News discussions