Novel Reasoning Effort Control Scheme Analyzed: Graded GRPO Training
stochasticchasm · x · 2026-09-11
stochasticchasm analyzes a novel reasoning effort control scheme, likening it to OpenAI's "juice" value for fine-grained control: each GRPO group samples at one effort level while each task samples at multiple levels, with an "Lnorm" reference length similar to k3's calibrated per-problem lengths — and suspects batch-invariant inference explains v4's throughput gains.
Related event: Leaked DeepSeek V4.1 Flash Tech Report Reveals KV Cache-Obsessed Flagship(44 posts)→
More from Models
- MissionReport benchmark: only 3 of 540 AI-generated government reports pass all criteria — Exp_Mark · 2026-10-27
- Local beats cloud: Qwen 3.8 FN handles private student data as well as frontier models — dr_lm · 2026-10-27
- Aidan Gomez: model's CoT said no tool calls right before making them — aidangomez · 2026-10-27
- Nat Lambert: Cohere's model is narrow RL training on internal infra ill-suited for open weights — natolambert · 2026-10-27
- OpenAI teases DOTS is coming to Europe 'soon' — mark_k · 2026-10-27
- Clifford Pickover challenges AI to decipher a message he encoded in Norse runes — pickover · 2026-10-27