Skip to content

A little signal. Less noise.

DISCOVER

Catalog Categories

CATEGORIES

Categories will appear when the source is connected.

More possibility. Less searching.

A considered collection of AI tools, trials, and opportunities.

RADAR / PostgreSQLConnection pending

ExploreCatalog

A little signal. Less noise.

Loading the catalog…

RADARGood tools open new possibilities.Explore thoughtfully. Create freely.

Skip to content

A little signal. Less noise.

DISCOVER

Catalog Categories

CATEGORIES

Categories will appear when the source is connected.

More possibility. Less searching.

A considered collection of AI tools, trials, and opportunities.

RADAR / PostgreSQLStorage available

ExploreOffer details

A little signal. Less noise.

Back to catalog

注

注意力不是全连接层换名字:多头自注意力的张量实验

掘金

About this offer

多头自注意力常被压缩成几行公式,于是维度、缩放和掩码稍有变化就难以定位错误。本文用一份不依赖深度学习框架的 Python 实验,把投影、分头、打分、稳定 Softmax、因果掩码和合并输出逐步展开,并

Offer details

Provider: 掘金
Source: juejin-rss
Score: 56.9
Confidence: 54%

Source evidence

What RADAR observed and classified to build this opportunity. It is what the source published, not a verification that the offer is still active.

JuejinObserved: Sep 29, 2026Source reliability: 60%
注意力不是全连接层换名字:多头自注意力的张量实验. 多头自注意力常被压缩成几行公式,于是维度、缩放和掩码稍有变化就难以定位错误。本文用一份不依赖深度学习框架的 Python 实验,把投影、分头、打分、稳定 Softmax、因果掩码和合并输出逐步展开,并
Open source

注意力不是全连接层换名字:多头自注意力的张量实验

Opens an external website. Availability and terms may change.

Real offers. Useful details.

The catalog shows persisted RADAR opportunities. Storage availability does not mean sources are verified or offers are active.

RADARGood tools open new possibilities.Explore thoughtfully. Create freely.