Toolcraft 创作工坊

开始创作

Academic Blueprint · 15 PAGES

Attention Is All You Need

Transformer 原论文深读 — 蓝图图纸风,结构图 + 公式 + 表格,带备注与动画,Quick 路线生成 (2026-08 生成;2026-09 按当前规范适配表达并重新打包,幻灯片内容未重生成)

下载完整 PPTX 查看原始作品 ↗

本作品来自 MIT 许可的 PPT Master Examples。下方展示全部 15 页；点击页面可放大浏览。

第 1 页：Attention Is All You Need — 01 / 15Attention Is All You Need

第 2 页：What this drawing set covers — 02 / 15What this drawing set covers

第 3 页：Before 2017, models paid for word order with time — 03 / 15Before 2017, models paid for word order with time

第 4 页：Keep attention, drop everything else — 04 / 15Keep attention, drop everything else

第 5 页：Six identical layers on each side — 05 / 15Six identical layers on each side

第 6 页：Scaled dot-product attention — 06 / 15Scaled dot-product attention

第 7 页：Multi-head attention: eight views at once — 07 / 15Multi-head attention: eight views at once

第 8 页：The same block is used in three places — 08 / 15The same block is used in three places

第 9 页：Injecting order without recurrence — 09 / 15Injecting order without recurrence

第 10 页：Three costs, four layer types (Table 1) — 10 / 15Three costs, four layer types (Table 1)

第 11 页：The spec sheet: data, hardware, schedule — 11 / 15The spec sheet: data, hardware, schedule

第 12 页：State of the art on both WMT 2014 tasks — 12 / 15State of the art on both WMT 2014 tasks

第 13 页：What the ablations taught (Table 3) — 13 / 15What the ablations taught (Table 3)

第 14 页：What was built on it — 14 / 15What was built on it

第 15 页：Attention is all you need — the rest was engineering — 15 / 15Attention is all you need — the rest was engineering

← 返回全部作品下载完整 PPTX