TRACE: Rollout-Guided Quantization-Aware Training for FP4 Reinforcement Learning of MoE Language Models

arXiv:2610.07767 · cs.LG, cs.CL · Submitted 2026-10-06 · Read on arXiv

cs.LG, cs.CL

Submitted: 2026-10-06

Updated: 2026-10-09

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Terminology

• Quantization — 1× in this paper · explained in 12 episodes
• Quantization-Aware Training — 1× in this paper · explained in 3 episodes
• Reinforcement Learning — 1× in this paper · explained in 4 episodes
• Trace — 1× in this paper · explained in 2 episodes

Sources

• Four Over Six: More Accurate NVFP4 Quantization with Adaptive Block Scaling
• PR2: Predictive Routing Replay for MoE-Based LLM Reinforcement Learning
• QeRL: Beyond Efficiency -- Quantization-enhanced Reinforcement Learning for LLMs
• DeepSWE: Measuring Frontier Coding Agents on Original, Long-Horizon Engineering Tasks
• QuRL: Efficient Reinforcement Learning with Quantized Rollout
• Stabilizing MoE Reinforcement Learning by Aligning Training and Inference Routers
• HiFloat4 Format for End-To-End Reinforcement Learning Post-Training of Large Language Models
• Score Centering Stabilizes Off-policy Reinforcement Learning
• Terminal-Bench: Benchmarking Agents on Hard, Realistic Tasks in Command Line Interfaces
• GDPval: Evaluating AI Model Performance on Real-World Economically Valuable Tasks
• On the Design of Qwen3.8-Next Architecture: Evaluation, Efficiency, and Training Stability
• DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models
• Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism
• Reinforcement Learning Optimization for Large-Scale Learning: An Efficient and User-Friendly Scaling Library
• The Internet of Things in the Era of Generative AI: Vision and Challenges
• Jet-RL: Enabling On-Policy FP8 Reinforcement Learning with Unified Training and Rollout Precision Flow
• H-Scale: Hessian-Guided Scale Refinement for NVFP4 Sub-Byte LLM Inference
• A Survey of Reinforcement Learning for Large Reasoning Models
• A Survey of Large Language Models
• Stabilizing Reinforcement Learning with LLMs: Formulation and Practices

Related papers

• Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
• AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
• Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
• Convergence issues in Relational Concept Analysis based on AOC-posets
• Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
• Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks