量化递归推理模型
Quantizing Recursive Reasoning Models
浏览论文内容
中文总结 AI 辅助
研究递归推理模型量化问题，发现其因激活缩放粒度致精度崩溃，提出用逐块缩放恢复转换，应用MXInt4格式，该格式在任务中与浮点格式有竞争力，克服架构量化敏感性弱点，还能转移到ARC - AGI基准测试。
中文摘要 AI 辅助
递归推理模型通过在多个细化步骤中应用紧凑、权重绑定的模块来解决难题。由于这些模块被多次重用，对其进行量化会产生一个独特的动态问题：每一步都会产生量化误差。虽然8位量化（整数或浮点数）能保持精度，但转换为张量级4位格式会导致系统偏差累积。这使得数独的精确解精度从84.1%灾难性地降至0.0%（只有约25%的单元格正确）。本文表明这种崩溃是由激活缩放粒度而非位宽或数字格式导致的。关键在于采用逐块缩放可完全恢复转换。为此，将MXInt4（一种逐块整数激活格式）应用于递归推理模型。在我们的任务中，它与逐块浮点格式具有竞争力，同时保持整数元素和2的幂次块尺度。最后，递归深度和重用会调节量化敏感性，我们测试的最深架构（EqR平衡模型）最敏感。然而，逐块缩放克服了这一弱点，在这些架构中保持稳健，并能转移到开放式ARC - AGI基准测试中。
英文摘要
Recursive reasoning models solve hard puzzles by applying compact, weight-tied blocks over many refinement steps. Because these blocks are reused many times, quantizing them creates a unique dynamical problem: the quantization error is incurred at every step. While 8-bit quantization (integer or float) preserves accuracy, moving to a per-tensor 4-bit format causes a systematic bias to accumulate. The ensuing drift catastrophically degrades exact-solution accuracy on Sudoku from 84.1% to 0.0% (only ~25% of cells correct). In this work, we show that this collapse is caused by activation-scaling granularity rather than bit-width or number format. Crucially, moving to per-block scaling completely restores the transition. To implement this, we apply MXInt4, a blockwise integer activation format, to recursive reasoning models. It is competitive with blockwise float formats on our tasks, while keeping integer elements and power-of-two block scales. Finally, recursion depth and reuse modulate quantization sensitivity, with the deepest architecture we test (the EqR equilibrium model) the most sensitive. Yet blockwise scaling overcomes this vulnerability, staying robust across these architectures and transferring to the open-ended ARC-AGI benchmark.
发表机构
- Integrated Systems Laboratory, ETH Zürich（集成系统实验室，瑞士苏黎世联邦理工学院）
机构由 AI 辅助整理，请以论文原文为准。
