AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

C-GRPO: Conformal Group Relative Policy Optimization

Arya Fayyazi, Seyedarmin Azizi, Mehdi Kamal, Massoud Pedram

NeurIPS 2026 · Reinforcement learning › RL from feedback · approach: GRPO stability and variants · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes C-GRPO, a GRPO variant that adapts the number of on-policy rollouts per prompt using a conformal stopping rule with a finite-sample calibration guarantee.

Problem

Standard GRPO uses a fixed group size, wasting compute on easy prompts and under-exploring hard ones; existing adaptive-budget methods use heuristic stopping rules without finite-sample calibration guarantees.

Approach

Replaces heuristic adaptive rollout stopping with a conformal calibration procedure that provides a finite-sample guarantee on the stopping decision within GRPO training.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace fixed G=32 GRPO rollouts with conformal adaptive budgetsGRPO on Qwen2.5-7B-Instruct · six benchmarkstraining rollouts: 49-70% fewer rollouts at matched accuracyImproved
use C-GRPO instead of static GRPOQwen2.5-7B-Instruct · GSM8Kpass@1 (%): 54.8 -> 57.2No clear effect
apply C-GRPO to a stronger base modelLlama-3.1-8B-Instruct · GSM8Krollout savings: 68% -> 82% (mean rollouts 10.4 -> 5.7)Improved
train fixed-G GRPO at matched compute G=10 (ablation)GRPO · GSM8Kpass@1 (%): 53.1 vs 57.2 for C-GRPOHurt
use fixed miscoverage 0.05 instead of auto-delta (ablation)C-GRPO · GSM8Kmean training rollouts: 10.4 -> 29.5; pass@1 57.2 -> 54.5Hurt
add automatic miscoverage selection (ablation)C-GRPO · GSM8Kpass@1 (%): 54.5 -> 57.2 vs fixed 0.05Improved
recalibrate every 25 steps instead of 50 (ablation)C-GRPO · GSM8Kpass@1 (%): 57.2 -> 57.3; +12 overheadNo clear effect
never recalibrate (static thresholds from step 0) (ablation)C-GRPO · GSM8Kmean training rollouts: 10.4 -> 18.2Hurt
add periodic recalibration at 100-step interval versus 50 (ablation)C-GRPO · GSM8Kpass@1 (%): 57.2 -> 55.8Hurt
use coarse budget grid {2,8,32} instead of {2,4,8,16,32} (ablation)C-GRPO · MATH-500pass@16 (%): 71.2 -> 69.5Hurt
use finer budget grid including k=1 (ablation)C-GRPO · MATH-500pass@16 (%): 71.2 -> 70.8No clear effect
train C-GRPO for 1,000 GSM8K stepsC-GRPO · GSM8Kpass@1 (%): 52.1 -> 64.5; mean k 18.2 -> 4.2Improved

Experimental setup

Baselines: GRPO adaptive-budget GRPO variants

Builds on: GRPO

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy