← 研踪 PaperTrailarXiv:2611.26106 · v1
AscMLA: Unlocking FP4 Efficiency for Long-Context MLA Decoding through Asymmetric Scaling
首次提交 2026-11-19 21:37
版本更新 2026-11-19 21:37
雷达观测 2026-11-20 16:53
摘要
Long-context decoding repeatedly reads a growing KV cache. FP4 reduces cache storage and traffic but needs block scaling to handle its limited dynamic range. Applying FP4 to multi-head latent attention (MLA) requires reconciling different K/V scaling requirements on a shared latent. Existing approaches address this mismatch through dequantization or separate encodings, adding execution or storage overhead. We present AscMLA, which combines asymmetric K/V scaling with pipelined scale delivery to process a single FP4 payload directly. AscMLA uses 38.2% less cache memory than FP8 and achieves geometric-mean speedups of 1.35x over TRT-LLM's FP8 and 1.26x over CAKE's NVFP4 MLA kernels.
观测到的版本
相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。
v1 · 2026-11-19 21:37 · 最新观测版本
AscMLA: Unlocking FP4 Efficiency for Long-Context MLA Decoding through Asymmetric Scaling
Long-context decoding repeatedly reads a growing KV cache. FP4 reduces cache storage and traffic but needs block scaling to handle its limited dynamic range. Applying FP4 to multi-head latent attention (MLA) requires reconciling different K/V scaling requirements on a shared latent. Existing approaches address this mismatch through dequantization or separate encodings, adding execution or storage overhead. We present AscMLA, which combines asymmetric K/V scaling with pipelined scale delivery to process a single FP4 payload directly. AscMLA uses 38.2% less cache memory than FP8 and achieves geometric-mean speedups of 1.35x over TRT-LLM's FP8 and 1.26x over CAKE's NVFP4 MLA kernels.
查看此版本解读 →查看此版本原文 ↗