PulseAugur
中
实时 22:26:15
English(EN) How can a small language model think longer without getting bigger? Looped transformers run one shared block of layers several times per token. NVIDIA's 4B Dory

NVIDIA的4B Dory使用循环Transformer来增强小型LLM的推理能力

研究人员正在探索在不增加体量的情况下增强小型语言模型推理能力的方法。一种技术涉及循环Transformer,它对每个token重复处理一个共享的层块。NVIDIA的4B Dory模型通过将96层通过一个20层的核心运行四次来展示这一点,保持权重不变但每次传递都会改变隐藏状态。这种方法以延迟为代价增加了计算深度。 AI

影响 这种方法可以实现更强大的小型模型,从而减少对海量计算资源的需求。

排序理由 该项目描述了一种新颖的语言模型架构方法,属于研究范畴。[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 →

NVIDIA的4B Dory使用循环Transformer来增强小型LLM的推理能力

本文如何被排名

Signal score
16 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
该项目描述了一种新颖的语言模型架构方法,属于研究范畴。[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) · michalpiszczek ·

    小型语言模型如何在不增大体量的情况下思考更长?循环Transformer对每个token运行一次共享层块。NVIDIA的4B Dory

    How can a small language model think longer without getting bigger? Looped transformers run one shared block of layers several times per token. NVIDIA's 4B Dory stores 36 layers and executes 96 when its 20-layer core runs 4 times. The weights never change; only the hidden state d…