Articles tagged with rlvr | Vitor Sousa

Vitor Sousa

Articles tagged rlvr

Back to blog
A Transformer from Raw Tensors: What 112 Tests Taught Me The model-foundations close-out of Learning with Verifiable Rewards: a decoder-only transformer built from raw tensors — attention with RoPE, RMSNorm, SwiGLU, an incremental KV-cache, four sampling modes — and the three properties that are easiest to get silently wrong. A case study in verification, not assembly. ~8 min
After GDPO: A Map of the GRPO Family Part 5 GRPO didn't fragment into a zoo. It exposed a handful of design axes — token vs. sequence granularity, clip width, length and std normalization, KL, advantage normalization, sampling — and every 'new' algorithm is a different point in that space. A practitioner's map of Dr. GRPO, DAPO, GSPO, GDPO, and the frontier, with a decision guide for when to reach for each. ~18 min