Search papers, labs, and topics across Lattice.
This paper introduces TRACE, a novel FP4 quantization framework for reinforcement learning of Mixture-of-Experts (MoE) language models, which directly minimizes the discrepancy between training and rollout quantization paths. By employing rollout-guided quantization-aware training and an efficient caching scheme, TRACE significantly reduces computational and memory overhead during RL training. Experimental results show TRACE achieves RL performance comparable to BF16 rollouts with substantial speedups and superior final FP4 performance.
FP4 RL for MoE LLMs can now match BF16 rollout quality with 5.4x speedup by directly aligning train-rollout quantization.
Reinforcement learning (RL) for post-training large language models (LLMs) incurs substantial computation and memory overhead during rollout generation, which motivates low-precision rollout for efficient RL training. However, existing FP4 RL methods suffer from a key limitation: they primarily optimize quantization accuracy on the training and rollout paths independently rather than directly reducing the discrepancy between the two quantized execution paths. In this work, we propose TRACE (Train-Rollout Quantization Alignment via Compact GuidancE), an FP4 quantization framework for RL training of Mixture-of-Experts (MoE) language models that addresses the limitation of existing FP4 RL methods. TRACE incorporates rollout-guided quantization-aware training that uses rollout-side quantization outcomes to guide training-side FP4 rounding decisions, directly reducing train-rollout discrepancy. Moreover, TRACE adopts an efficient quantization-information caching scheme that selectively retains mantissa and scale information from deeper layers to reduce the storage and communication overhead introduced by rollout guidance. We evaluate TRACE on four large-scale MoE language models across reasoning, coding, and long-horizon RL tasks. Our results demonstrate that TRACE enables joint FP4 weight/activation and FP4 KV-cache rollout with RL performance comparable to BF16 rollout, while achieving up to 5.4xrollout speedup and strong final FP4 performance compared with post-hoc FP4 quantization of BF16-trained policies.
