# TRIAGE: Direction-Aware Mismatch Stabilization of Native NVFP4 Reinforcement Learning

arXiv: https://arxiv.org/abs/2610.07043 · HTML: https://mlsanity.com/paper/2610.07043/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in RL training of language and agent models, one of 15 directions in Deep reinforcement learning methods.
- Builds on 19 of its 24 references that are in the map.
- First posted on arXiv 5 Oct 2026.

## Placed in

- Deep reinforcement learning methods / [RL training of language and agent models](https://mlsanity.com/deeprl/rl-training-of-language-and-agent-models/)

## Builds on (most cited first, 19 of 19)

- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](https://mlsanity.com/paper/2501.12948/) (2025)
- [Let's Verify Step by Step](https://mlsanity.com/paper/2305.20050/) (2023)
- [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://mlsanity.com/paper/1909.08053/) (2019)
- [Pretraining Large Language Models with NVFP4](https://mlsanity.com/paper/2509.25149/) (2025)
- [Quantization Meets Reasoning: Exploring LLM Low-Bit Quantization Degradation for Mathematical Reasoning](https://mlsanity.com/paper/2501.03035/) (2025)
- [Composer 2 Technical Report](https://mlsanity.com/paper/2603.24477/) (2026)
- [Quartet II: Accurate LLM Pre-Training in NVFP4 by Improved Unbiased Gradient Estimation](https://mlsanity.com/paper/2601.22813/) (2026)
- [Scaling Law for Quantization-Aware Training](https://mlsanity.com/paper/2505.14302/) (2025)
- [QeRL: Beyond Efficiency -- Quantization-enhanced Reinforcement Learning for LLMs](https://mlsanity.com/paper/2510.11696/) (2025)
- [FP8-RL: A Practical and Stable Low-Precision Stack for LLM Reinforcement Learning](https://mlsanity.com/paper/2601.18150/) (2026)
- [QuRL: Efficient Reinforcement Learning with Quantized Rollout](https://mlsanity.com/paper/2602.13953/) (2026)
- [QaRL: Rollout-Aligned Quantization-Aware RL for Fast and Stable Training under Training--Inference Mismatch](https://mlsanity.com/paper/2604.07853/) (2026)
- [Diagnosing Training Inference Mismatch in LLM Reinforcement Learning via a Zero-Mismatch Reference](https://mlsanity.com/paper/2605.14220/) (2026)
- [SSPO: Subsentence-level Policy Optimization](https://mlsanity.com/paper/2511.04256/) (2025)
- [QUADS: Stabilizing NVFP4 Reinforcement Learning for MoE via QUantization-error Alignment across Dual Sides](https://mlsanity.com/paper/2607.15810/) (2026)
- [InfiR : Crafting Effective Small Language Models and Multimodal Small Language Models in Reasoning](https://mlsanity.com/paper/2502.11573/) (2025)
- [ACRL: Adaptive Control of Training-Inference Discrepancy for Stable Reinforcement Learning](https://mlsanity.com/paper/2607.24062/) (2026)

## Abstract (arXiv)

Low-precision execution can substantially accelerate reinforcement learning (RL) for large language models, but discrepancies between learner and sampler execution can destabilize policy optimization. In this paper, we characterize the interaction between mismatch and the policy-gradient direction, distinguishing locally amplifying from contracting update contributions that mismatch magnitude alone cannot identify. In native NVFP4 runs, we observe an early imbalance between the two amplifying regions, favoring negative-advantage, negative-gap updates. Their tail tokens become concentrated in a small fraction of response segments before mismatch spreads globally. Motivated by these findings, we introduce TRIAGE, a direction-aware stabilization method that uses segment-level diagnosis to selectively rebalance policy-gradient updates and applies bounded repair to residual severe mismatch. TRIAGE modifies the optimization objective while retaining native NVFP4 weight-and activation 4-bit (W4A4) forward execution on both the sampler and learner. Experiments on Qwen3-4B and Qwen3-30B-A3B show stable optimization throughout the evaluated training horizon and achieve full precision level performance across five mathematical reasoning benchmarks, while native NVFP4 with TRIAGE provides up to 2.3x higher rollout throughput than BF16.