English(EN) Mixture-of-Depths: a per-block router keeps top-k tokens, the rest skip the residual — dynamic depth at a fixed FLOP budget

Mixture-of-Depths 动态分配 Transformer 计算量，节省约 50% 的 FLOPs

作者 PulseAugur 编辑部 · [1 个来源] · 2026-07-28 15:20

研究人员开发了一种名为 Mixture-of-Depths (MoD) 的新颖技术，该技术可以动态分配 Transformer 模型内的计算资源。通过在每个块中引入一个小型路由器，MoD 仅识别和处理最重要的 token，允许不太关键的 token 跳过计算并沿残差连接传递。这种方法通过在固定的总预算下为每个 token 动态分配深度，显著减少了 FLOPs，以大约 50% 的计算量实现了与密集模型相当的性能。 AI

影响这项技术可以通过降低计算成本而不牺牲性能，从而实现更高效的 LLM 训练和推理。

排序理由详细介绍新模型架构技术的论文。[lever_c_demoted from research: ic=1 ai=1.0]

在 dev.to — LLM tag 阅读 →

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

Mixture-of-Depths 动态分配 Transformer 计算量，节省约 50% 的 FLOPs

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; 详细介绍新模型架构技术的论文。[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: paper, 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: 72 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]

dev.to — LLM tag TIER_1 English(EN) · Devanshu Biswas · 2026-07-28 15:20

Mixture-of-Depths：每块路由器保留前k个token，其余跳过残差连接——固定FLOP预算下的动态深度

A normal transformer forces every token through every layer, so the word "the" costs exactly as much compute as "photosynthesis." That's obviously wasteful, and Mixture-of-Depths (DeepMind, 2024) is the fix that finally made the waste addressable to me. It ad…

本文如何被排名

报道来源 [1]

Mixture-of-Depths：每块路由器保留前k个token，其余跳过残差连接——固定FLOP预算下的动态深度

相关实体

相关话题