English(EN) Tile-Level Activation Overlap for Efficient LLM Inference

新内核通过融合 SwiGLU 激活来提高大语言模型推理速度

作者 PulseAugur 编辑部 · [1 个来源] · 2026-07-07 04:00

研究人员开发了新的技术，通过在瓦片级别将 SwiGLU 激活函数直接融合到 GEMM 操作中，来加速大语言模型 (LLM) 的推理。这些方法使用为 NVIDIA H100 GPU 定制的 CUTLASS 内核实现，显著减少了中间张量物化的相关开销。在 Qwen 2.5 模型上的评估显示，与标准的 PyTorch 实现相比，速度提升高达 2.47 倍，实现了更高的峰值 BF16 利用率，并展示了优于 cuBLAS 等现有库的数值性能。 AI

影响这项研究可能有助于在大语言模型在生产环境中更高效的部署和更快的响应时间。

排序理由详细介绍一种新颖技术方法以提高大语言模型推理效率的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.LG 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 0 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 详细介绍一种新颖技术方法以提高大语言模型推理效率的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: infra, paper; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: 93 days old; Aged out of breaking-news scoring windows; ranking reflects the durable signal from the full source set.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.LG TIER_1 English(EN) · Abhinav Jangda, Tyler Sorensen, Sebastian Burckhardt, Jianlan YE, Chaoyin Li, Atul Gupta · 2026-07-07 04:00

面向高效LLM推理的瓦片级激活重叠

arXiv:2607.02521v1 Announce Type: cross Abstract: SwiGLU is the dominant MLP activation in modern large language models, yet its intermediate tensor materialization costs 9-37% of MLP execution time. We present two complementary CUTLASS-based SM90 kernels that fuse SwiGLU into Ge…

本文如何被排名

报道来源 [1]

面向高效LLM推理的瓦片级激活重叠

相关实体

相关话题