Search papers, labs, and topics across Lattice.
This paper identifies the critical issue of training-inference mismatch in reinforcement learning for large language models (LLMs), which leads to inconsistent performance and instability during training. The authors propose a novel optimization objective, Monotonic Inference Policy Improvement (MIPI), alongside a two-step framework called Monotonic Inference Policy Update (MIPU) that enhances training stability and reasoning performance by aligning training updates with inference policies. Experimental results demonstrate that MIPU significantly improves average reasoning performance and mitigates training fragility across different model scales under high mismatch conditions.
Training updates that improve performance in LLMs can actually degrade inference quality鈥攗nless you use the new Monotonic Inference Policy Update framework.
Reinforcement learning (RL) has gained growing attention in large language model (LLM) post-training, yet RL training remains fragile and can suffer from instability or collapse. One vital cause is training-inference mismatch: LLM adopts separate inference and training engines for generation efficiency and training precision, which in practice exhibits inconsistent probabilities for the same trajectories on training and inference sides, even with synchronized model parameters. This naturally induces a special type of off-policyness ever existing and poisoning the training. Prior works have made various efforts in addressing the off-policyness to stabilize the training policies under the mismatch. In this paper, we point out the objective misalignment neglected by existing works that an effective update to the policy in the training engine not necessarily ensures the improvement of the inference policy, i.e., the one used in deployment. To this end, we propose a new policy optimization objective for LLM RL, named Monotonic Inference Policy Improvement (MIPI). Following this principle, we introduce Monotonic Inference Policy Update (MIPU), a two-step LLM RL framework that constructs sampler-referenced candidate updates and selectively accepts synchronized candidates using an inference-side gap proxy. Experiments conducted on two model scales under high mismatch show that MIPU improves average reasoning performance and training stability.
