做推理模型量化和部署的团队终于有了针对 2-bit 失败模式的系统解法——不是简单降精度，而是用 FP16 规划和循环救援来修复生成过程，Qwen3 用户可以直接复现并提升准确率。
该研究揭示了大型推理模型在极端低比特（2-bit）量化推理时，并非单纯降低答案准确率，而是产生更长的推理轨迹，包括重复循环、预算耗尽、延迟决策和未闭合推理段，导致端到端速度不升反降。作者针对 Qwen3-8B 和 Qwen3-32B 模型，提出了两种轻量级控制方法：FP16 规划（为 2-bit 模型提供短的高精度大纲）和循环救援（检测重复轨迹并回退或提前提交答案）。在 MATH-500 上，循环救援将 Qwen3-8B 准确率从 17.2% 提升至 74.2%，规划加循环救援将 Qwen3-32B 从 65.0% 提升至 87.2%。研究表明，将低比特推理失败视为可控生成病理，通过轻量检测和选择性 FP16 支持，2-bit 推理可以恢复准确率并保持真实端到端加速。代码已开源。
Extreme Low-Bit Inference in Reasoning Models: Failure Modes and Targeted Recovery
Large Reasoning Models (LRMs) rely on long reasoning traces, making inference expensive. While low-bit quantization reduces per-token decoding cost, we show that aggressive 2-bit inference can fail to deliver end-to-end speedup because instability in the generation process inflates total token count. Instead of merely lowering answer accuracy, 2-bit quantization often produces much longer traces with repetitive loops, budget exhaustion, delayed commitment, and unclosed reasoning segments. We analyze full reasoning traces of Qwen3 reasoning models across mathematical and commonsense benchmarks and show that accuracy degradation is tightly linked to these process-level failures. To address them, we introduce two lightweight controls: FP16 planning, which gives the 2-bit model a short high-precision outline, and loop rescue, which detects repetitive traces and either commits to an earlier answer or falls back to FP16. On MATH-500, loop rescue improves Qwen3-8B accuracy from 17.2% to 74.2%, while planning plus loop rescue improves Qwen3-32B from 65.0% to 87.2%. Overall, our results show that extreme low-bit reasoning becomes practical when its failures are treated as controllable generation pathologies: with lightweight detection and selective FP16 support, 2-bit inference can recover accuracy while preserving real end-to-end speed. Our code is available at: https://github.com/brain-lab-research/quantized-reasoning.