PulseAugur
实时 00:39:54
English(EN) FusionML: Prefill, Not Decode - Mechanism and Boundaries of CPU+GPU Co-Execution on Unified-Memory Apple Silicon

FusionML 通过 CPU-GPU 协同执行提升 Apple Silicon 的 AI 推理性能

研究人员开发了 FusionML 系统,通过协同执行 CPU 和 GPU 操作来优化 Apple Silicon 上的 Transformer 推理。通过解决 MLX 调度器导致的串行化限制,FusionML 实现了 Transformer 预填充的行切分方法,使 Llama 模型提速 1.15 倍至 1.38 倍,Qwen2.5-7B 的首次 token 时间缩短 1.18 倍至 1.25 倍。该系统在多代 Apple Silicon 上的有效性得到了表征,研究结果表明解码操作并未受益,而某些训练场景或 Apple Neural Engine 会引入过高的开销。 AI

影响 优化 Apple Silicon 上的 AI 推理,可能提升设备端 AI 应用的性能。

排序理由 详细介绍在特定硬件上优化 AI 模型推理新方法的论文。[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 →

FusionML 通过 CPU-GPU 协同执行提升 Apple Silicon 的 AI 推理性能

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
详细介绍在特定硬件上优化 AI 模型推理新方法的论文。[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, model release
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
49 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]

  1. arXiv cs.LG TIER_1 English(EN) · Om Mohite ·

    FusionML:预填充而非解码——统一内存Apple Silicon上CPU+GPU协同执行的机制与边界

    arXiv:2607.22785v1 Announce Type: cross Abstract: Apple-Silicon SoCs share CPU, GPU, and Neural Engine over one unified memory system, raising the question of whether transformer inference can be accelerated by splitting single operators across units. Prior attempts, including ou…