aiqbench analysis
MXAttention addresses critical numerical instabilities in MXFP4 quantization for diffusion models by introducing a data-free scaling method and pre-normalization techniques. This approach successfully recovers nearly 95% of the quality gap between low-bit precision and full FP16 without requiring calibration datasets or search procedures. The technique offers a viable path to efficient inference for video generation tasks while maintaining high fidelity in generated frames.
MXAttention: Data-Free Optimal Scaling and Pre-Normalization Quantization for MXFP4 Attention
arXiv:2607.24377v1 Announce Type: cross Abstract: The quadratic cost of attention is a major bottleneck in diffusion-based video generation models. MXFP4 attention provides a promising path toward efficient inference, but direct MXFP4 quantization often degrades generation quality due to two numerical issues: the clipping-underflow trade-off from power-of-two scaling and the row-wise normalization error introduced in the softmax loop. We propose MXAttention, a data-free post-training quantization framework for MXFP4 attention. MXAttention introduces two components: Universal Optimal Scaling (UOS), which exploits the periodic structure of power-of-two microscaling to derive a distribution-independent optimal scaling boundary Qmax=7.25 without calibration or search, and Pre-Normalization Quantization (PNQ), which quantizes unnormalized softmax exponentials before row-wise summation to preserve normalization by construction. Experiments on Wan2.2 and HunyuanVideo show that MXAttention closes at least 95% of the VBench Imaging Quality gap between OCP MXFP4 and FP16, substantially improves frame-level similarity, and preserves FP16-level generation quality with less than 0.01 absolute degradation on all reported VBench metrics. MXAttention also achieves performance competitive with strong NVFP4-based baselines with negligible overhead when fused into the attention pipeline. The implementation is publicly available in MindIE-SD.
Abstract:The quadratic cost of attention is a major bottleneck in diffusion-based video generation models. MXFP4 attention provides a promising path toward efficient inference, but direct MXFP4 quantization often degrades generation quality due to two numerical issues: the clipping-underflow trade-off from power-of-two scaling and the row-wise normalization error introduced in the softmax loop. We propose MXAttention, a data-free post-training quantization framework for MXFP4 attention. MXAttention introduces two components: Universal Optimal Scaling (UOS), which exploits the periodic structure of power-of-two microscaling to derive a distribution-independent optimal scaling boundary Qmax=7.25 without calibration or search, and Pre-Normalization Quantization (PNQ), which quantizes unnormalized softmax exponentials before row-wise summation to preserve normalization by construction. Experiments on Wan2.2 and HunyuanVideo show that MXAttention closes at least 95% of the VBench Imaging Quality gap between OCP MXFP4 and FP16, substantially improves frame-level similarity, and preserves FP16-level generation quality with less than 0.01 absolute degradation on all reported VBench metrics. MXAttention also achieves performance competitive with strong NVFP4-based baselines with negligible overhead when fused into the attention pipeline. The implementation is publicly available in MindIE-SD.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
arXiv:2607.24377 [cs.LG]
(or arXiv:2607.24377v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2607.24377 arXiv-issued DOI via DataCite (pending registration)
Full text under the publisher's open licence · 203 words
Related coverage
DI Series: Serving Kimi-K3 MXFP4 Wide-EP Disaggregated on AMD Instinct MI300X and MI325X
AMD · Oct 8, 2026
[AINews] Agents disprove Polysemanticity; greatly improve Interpretability & Alignment
Latent Space · Oct 16, 2026
Optimizing ATOM and vLLM-ATOM for High-Interactivity Inference
AMD · Sep 1, 2026
GEM Training: How Meta Doubled the Efficiency of Its LLM-Scale Ads Foundation Model
Meta Engineering · Aug 3, 2026