phenomenon · first seen 5 Oct 2026
reinforcement learning from policy gradients with train-inference mismatch
Also: TIM, RL rollout mismatch, training-inference mismatch
Numerical discrepancy between the learner and sampler in RL for LLMs that can destabilize policy optimization.
Mentions per week
PapersCodeDiscussion
Earliest sources
- TRIAGE: Direction-Aware Mismatch Stabilization of Native NVFP4 Reinforcement LearningPapers · 5 Oct 2026 · Zhen Li, Shuai Zhang, Yanggan Gu, Yiming Zhang, Yang Yu, Mingfa Feng, Congkai Xi…
- On the Origin and Evolution of Training-Inference Mismatch in LLM Reinforcement LearningPapers · 29 Oct 2026 · Yu Li, Yufa Zhou, Tian Lan
- FourSight: Stable FP4 RL Post-Training via Sequence Filtering and Sparse Channel CompensationPapers · 3 Nov 2026 · Yining Wei, Tianyuan Wu, Gu Gong, Yuechen Tao, Yinghao Yu, Xiaowen Chu, Wei Wang
- Inferact/zero-mismatch-rlCode · 3 Nov 2026 · Inferact · 44 stars
- How Much Training-Inference Mismatch Can On-Policy Distillation Tolerate?Papers · 8 Nov 2026 · Cong Yu, Ziting Wang, Songlin Jiang, Sitong Zhang, Chengliang Chai, Youhe Jiang,…
Latest sources
- How Much Training-Inference Mismatch Can On-Policy Distillation Tolerate?Papers · 8 Nov 2026 · Cong Yu, Ziting Wang, Songlin Jiang, Sitong Zhang, Chengliang Chai, Youhe Jiang,…
- Inferact/zero-mismatch-rlCode · 3 Nov 2026 · Inferact · 44 stars
- FourSight: Stable FP4 RL Post-Training via Sequence Filtering and Sparse Channel CompensationPapers · 3 Nov 2026 · Yining Wei, Tianyuan Wu, Gu Gong, Yuechen Tao, Yinghao Yu, Xiaowen Chu, Wei Wang
- On the Origin and Evolution of Training-Inference Mismatch in LLM Reinforcement LearningPapers · 29 Oct 2026 · Yu Li, Yufa Zhou, Tian Lan
- TRIAGE: Direction-Aware Mismatch Stabilization of Native NVFP4 Reinforcement LearningPapers · 5 Oct 2026 · Zhen Li, Shuai Zhang, Yanggan Gu, Yiming Zhang, Yang Yu, Mingfa Feng, Congkai Xi…