PulseAugur
中
实时 04:15:18
English(EN) b11430: hexagon: matmul and flash-atten scalability updates (#29974)

llama.cpp 优化 hexagon 架构以加速 AI 模型推理

llama.cpp 项目发布了更新 b11430,专注于 hexagon 架构的性能改进。此次更新为 flash-attention 引入了头并行分区,允许每个核心处理 KV 缓存的特定头分片。这一改变显著提升了 Qwen3-0.6B 和 Llama 3.2:3b 等特定模型的性能,分别测量到高达 58% 和 49% 的提升。更新还包括了受 GGML_HEXAGON_FA_HEAD_SPLIT 标志控制的矩阵乘法和多设备场景的各种优化。 AI

影响 提高了特定硬件架构上 AI 模型的推理速度。

排序理由 开源项目的软件更新,专注于性能优化。

在 llama.cpp — Releases 阅读 →

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

llama.cpp 优化 hexagon 架构以加速 AI 模型推理

本文如何被排名

Signal score
2 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
开源项目的软件更新,专注于性能优化。
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
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
Same-day
Cluster formed today. Ranking reflects the current source set at time of score.

完整方法见我们的编辑标准。

报道来源 [1]

  1. llama.cpp — Releases TIER_1 English(EN) · max-krasnyansky ·

    b11430: hexagon: matmul 和 flash-atten 可扩展性更新 (#29974)

    <ul> <li>hexagon: head-parallel flash_attn partitioning for row-split multicore</li> </ul> <p>In row-split mode each core computes its output row shard of every<br /> MUL_MAT, but flash_attn was previously partitioning by Q tokens<br /> (flat qrow split) instead of by heads. This…