Squeeze the Soaked Sponge: Efficient Off-policy RFT for Large Language Model

Jan 5, 2026·
Jing Liang
*
,
Jinyi Liu
Jinyi Liu
*
,
Yi Ma
*
,
Hongyao Tang
,
Yan Zheng
,
Shuyue Hu
,
Lei Bai
,
Jianye Hao
· 0 min read
Type
Publication
The Fourteenth International Conference on Learning Representations (ICLR 2026 Poster)

Overview

ReMix brings off-policy reinforcement finetuning to LLM post-training by reusing rollout data from past policies, dramatically reducing training cost while staying competitive on math reasoning benchmarks.

Venue. The Fourteenth International Conference on Learning Representations (ICLR 2026 Poster)

LLM Post-Training (RL Tuning) LLM Post-Training RLVR Off-Policy
Jinyi Liu
Authors
Jinyi Liu
Ph.D. Candidate Reinforcement Learning and LLM Systems
Back to Publications

More Publications

CellAgent: LLM-Driven Multi-Agent Framework for Natural Language-Based Single-Cell Analysis Jan 2026 From Seeing to Doing: Bridging Reasoning and Decision for Robotic Manipulation Jan 2026 Benchmarking Continual Agent Memory for Online Learning, Transfer, and Forgetting Jan 2026