4-bit权重与激活量化实现视频生成模型高效推理
W4A4 Quantization for Inference on Wan2.2-I2V-A14B
- 融合混合精度与通道平滑技术，处理稀疏异常激活
- 在多数指标上仅比全精度低2-3.5%，运动更流畅
- 适合部署大模型视频生成任务的轻量化推理
我们提交了参加ICME 2026低比特量化挑战赛子挑战1：基于HiF4或MXFP4格式的W4A4量化推理。目标是在万像模型Wan2.2-I2V-A14B上实现4比特权重与4比特激活的推理。借鉴大语言模型量化思想，采用MixQ式混合精度处理稀疏激活异常值、SmoothQuant式逐通道平滑，并结合块级HiF4打包用于前馈线性层。在OpenS2V-5M代表性数据批次上校准发现重尾激活通道；平滑操作在量化前重新分配动态范围；双分支GEMM保留异常通道的高精度，其余通道严格使用W4A4。在官方VBench I2V指标下，我们的方案在多数质量维度上仅比FP16低2-3.5%，且提升运动流畅性，显著优于原生HiFloat4基线（相对下降约5%）。
原文摘要 · Abstract (English)
We summarize our submission to Sub-Challenge 1: W4A4 Quantization for Inference (HiF4 / MXFP4) of the ICME 2026 Low-Bit-width Large-Model Quantization Challenge. The sub-challenge targets 4-bit weight and 4-bit activation inference on Wan-AI/Wan2.2-I2V-A14B under HiF4 or MXFP4 numerical formats. We adapt two complementary ideas from LLM quantization, MixQ-style mixed precision for sparse activation outliers and SmoothQuant-style per-channel smoothing, together with block-wise HiF4 packing for Wan2.2 feed-forward linear layers. Calibration on representative OpenS2V-5M batches identifies heavy-tailed activation channels; smoothing rebalances dynamic range before W4A4 rounding; and a dual-branch GEMM preserves outlier columns in higher precision while the bulk of channels use strict W4A4. On official VBench I2V metrics, our pipeline stays within 2-3.5 percent of FP16 on most quality axes and improves motion smoothness, outperforming a native HiFloat4 baseline that degrades roughly 5 percent relative to FP16 across all reported scores.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。