6 papers
Trust Region Masking for Long-Horizon LLM Reinforcement Learning
Yingru Li, Jiacai Liu, Jiawei Xu +4
Policy gradient methods for Large Language Models optimize a policy via a surrogate objective computed from samples of a rollout policy . However, modern…
The Optimal Token Baseline: Variance Reduction for Long-Horizon LLM-RL
Yingru Li, Jiawei Xu, Ziniu Li +10
Reinforcement Learning (RL) for Large Language Models (LLMs) often suffers from training collapse in long-horizon tasks due to exploding gradient variance. To mitigate this, a base…
Dynamic Vocabulary Pruning: Stable LLM-RL by Taming the Tail
Yingru Li, Jiawei Xu, Jiacai Liu +6
Reinforcement Learning (RL) for Large Language Models (LLMs) faces a fundamental tension: the numerical divergence between high-throughput inference engines and numerically precise…
Beyond Precision: Training-Inference Mismatch is an Optimization Problem and Simple LR Scheduling Fixes It
Yaxiang Zhang, Yingru Li, Jiacai Liu +4
Reinforcement Learning (RL) for training Large Language Models is notoriously unstable. While recent studies attribute this to "training inference mismatch stemming" from inconsist…
A Note on Hybrid Online Reinforcement and Imitation Learning for LLMs: Formulations and Algorithms
Yingru Li, Ziniu Li, Jiacai Liu
We present a unified framework for Large Language Model (LLM) fine-tuning that integrates Imitation Learning and Reinforcement Learning. By analyzing the gradient of a composite ob…
Harnessing Uncertainty: Entropy-Modulated Policy Gradients for Long-Horizon LLM Agents
Jiawei Wang, Jiacai Liu, Yuqian Fu +7
In long-horizon tasks, recent agents based on Large Language Models (LLMs) face a significant challenge that sparse, outcome-based rewards make it difficult to assign credit to int…