PulseAugur
中
实时 03:58:06
English(EN) ModelBest says its Lookahead Expert Scheduling runs a 35B-parameter mixture-of-experts model at over 20 tokens per second on an iPhone 18 Pro Max, activating ab

ModelBest 使 35B MoE 模型能在 iPhone 上以 20+ token/秒的速度运行

ModelBest 开发了一种名为 Lookahead Expert Scheduling 的技术,该技术允许一个拥有 350 亿参数的混合专家模型在 iPhone 18 Pro Max 上以超过每秒 20 个 token 的速度运行。该方法每个 token 激活约 30 亿个参数,专家权重按需从闪存加载。该公司还使用混合精度量化将模型的文件大小从 19.7GB 减少到 14.44GB,并且该代码已开源。 AI

影响 证明了在移动设备上运行大规模 MoE 模型的可行性,可能为更强大的设备端 AI 应用提供支持。

排序理由 将一种新颖的在消费设备上运行大型 MoE 模型的技术开源发布。[lever_c_demoted from research: ic=1 ai=1.0]

在 Mastodon — mastodon.social 阅读 →

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

ModelBest 使 35B MoE 模型能在 iPhone 上以 20+ token/秒的速度运行

本文如何被排名

Signal score
22 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
将一种新颖的在消费设备上运行大型 MoE 模型的技术开源发布。[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
model release, 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
Breaking (< 6h)
Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

  1. Mastodon — mastodon.social TIER_1 English(EN) · [email protected] ·

    ModelBest 表示其 Lookahead Expert Scheduling 在 iPhone 18 Pro Max 上运行 35B 参数的混合专家模型,速度超过每秒 20 个 token,激活 ab

    ModelBest says its Lookahead Expert Scheduling runs a 35B-parameter mixture-of-experts model at over 20 tokens per second on an iPhone 18 Pro Max, activating about 3B parameters per token. Expert weights load from flash on demand, and mixed-precision quantisation cut the weight f…