AI
Groupwise Agentic Grading and Advantage Redistribution for Code Agent RL
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 25/09/2026, 20:00:00 UTC
Briefing created: 29/09/2026, 22:18:35 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 29/09/2026, 10:18:24 UTC.
Briefings
This article presents GAGAR, a framework for quality-aware credit redistribution in code agent reinforcement learning. Unlike standard Group Relative Policy Optimization, GAGAR uses an SFT-trained agentic grader to rank test-passing trajectories within dynamic sampling groups. By downweighting lower-ranked candidates and rescaling advantages proportionally, the method preserves the original sum of rewards while shifting credit toward higher-quality implementations. The framework was evaluated at industrial scale using pre-RL SFT checkpoints of MiMo-V2.6-Flash (310B parameters) and MiMo-V2.6-Pro (1.02T parameters), resulting in improved code agent performance, reduced trajectory-length growth, and more stable training.
Key points
- GAGAR uses an SFT-trained agentic grader to rank test-passing trajectories within dynamic sampling groups before redistributing advantages.
Context and limitations
The framework was evaluated at industrial scale using pre-RL SFT checkpoints; mixed-task RL results are not yet reported.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 29/09/2026, 22:18:35 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeiha7xicrn6nils7ki7npsa57oj47r2hzgzku7gfb2cbit27gjaxr4