Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Foundation Models frameworkで画像分析
Search
Sponsored
·
SiteGround - Reliable hosting with speed, security, and support you can count on.
→
リョウ
July 18, 2026
Programming
110
1
Share
Embed
Copy iframe code
Copy JS code
Copy link
Start on current slide
Foundation Models frameworkで画像分析
WWDC26 Recap - Japan-\(region).swift
https://luma.com/x230s9i1
リョウ
July 18, 2026
More Decks by リョウ
See All by リョウ
オンデバイスAIとXcode
ryodeveloper
0
2.5k
アプリのユーザー体験を高めるAccessibility Modifier
ryodeveloper
0
140
Cursor
ryodeveloper
0
48
What’s new in Xcode 16
ryodeveloper
0
78
SwiftUI Introspect
ryodeveloper
0
150
ZStackとbackgroundの使い分け / Comparing ZStack and background
ryodeveloper
1
570
Other Decks in Programming
See All in Programming
5分で問診!Composer セキュリティ健康診断
codmoninc
0
150
PHPだって関数型したい 〜できること、できないこと〜 / fp-in-php
jsoizo
0
210
Claude Team Plan導入・ガイド
tk3fftk
0
190
フィードバックで育てるAI開発
kotaminato
1
110
継続モナドとリアクティブプログラミング
yukikurage
3
590
「正の参照」と 「負の導出」で組む ハーネスエンジニアリング
cottpan
1
140
ローカルLLMでどこまでコードが書けるか -拡張版 / How much code can be written on a local LLM Extended
kishida
12
4.8k
「なぜそう決めたのか」を残し続ける仕組み ― Notion AI カスタムエージェント × Slack連携による設計判断の自動記録 - NIKKEI Tech Talk #47
niftycorp
PRO
0
260
act1-costs.pdf
sumedhbala
0
230
エンジニアと一緒にテストコードの設計と実装を改善した話
mototakatsu
0
260
エンジニアにデザインハーネスを 〜デザインプロセスを規定するためのハーネス〜 / Design harness from an engineer's perspective
rkaga
2
1.4k
はてなアカウント基盤 State of the Union
cockscomb
1
1.3k
Featured
See All Featured
SEO Brein meetup: CTRL+C is not how to scale international SEO
lindahogenes
1
2.8k
The Web Performance Landscape in 2024 [PerfNow 2024]
tammyeverts
12
1.2k
Breaking role norms: Why Content Design is so much more than writing copy - Taylor Woolridge
uxyall
0
340
Jamie Indigo - Trashchat’s Guide to Black Boxes: Technical SEO Tactics for LLMs
techseoconnect
PRO
0
340
We Analyzed 250 Million AI Search Results: Here's What I Found
joshbly
1
1.5k
Reflections from 52 weeks, 52 projects
jeffersonlam
356
21k
Public Speaking Without Barfing On Your Shoes - THAT 2023
reverentgeek
1
460
brightonSEO & MeasureFest 2025 - Christian Goodrich - Winning strategies for Black Friday CRO & PPC
cargoodrich
3
750
Claude Code のすすめ
schroneko
67
230k
The untapped power of vector embeddings
frankvandijk
2
1.8k
SEOcharity - Dark patterns in SEO and UX: How to avoid them and build a more ethical web
sarafernandez
0
220
Leading Effective Engineering Teams in the AI Era
addyosmani
9
2.1k
Transcript
Ϧϣ @RyoDeveloper Foundation Models framework Ͱը૾ੳ WWDC26 Recap - Japan-\(region).swift
Ϧϣ • @RyoDeveloper • גࣜձࣾLuup iOS Engineer
Foundation Models framework Apple Intelligenceͷج൫ͱͳΔେنݴޠϞσϧΛ Swift API͔Βར༻Ͱ͖ΔϑϨʔϜϫʔΫ • ϓϥΠόγʔͷ֬อ •
ΦϑϥΠϯ࣮ߦՄೳ • OSʹΈࠐ·Ε͍ͯΔ • ແྉͰར༻Մೳ
WWDC26ͰͷFoundation Models frameworkͷΞοϓσʔτ • ը૾ೖྗ • Evaluations framework • ίΞϑϨʔϜϫʔΫͷΦʔϓϯιʔεԽ
• ৽͍͠ΦϯσόΠεϞσϧ • Private Cloud Compute • αʔυύʔςΟϞσϧରԠ
None
LUUPΞϓϦͰͷ ׆༻ ※PoCͱ࣮ͯͨ͠͠Πϝʔδ Ͱ͢
Foundation Models frameworkͷը૾ೖྗ let session = LanguageModelSession() let response =
try await session.respond { "͜ͷը૾ʹ͍ࣸͬͯΔं྆ʹΕ͕ͳ͍͔ڭ͑ͯɻ" Attachment(image) } let content = response.content ͜ͷं྆ͷલΧΰʹࠇ͍ϦϡοΫ͕ೖ͍ͬͯ·͢ɻ
Guided Generation @Generable struct ForgottenItemPresence { @Guide(description: "ं྆ʹΕ͕͍ࣸͬͯΔ͔") let hasForgottenItem:
Bool } let session = LanguageModelSession() let response = try await session.respond( generating: ForgottenItemPresence.self ) { "͜ͷը૾ʹ͍ࣸͬͯΔं྆ʹΕ͕ͳ͍͔ڭ͑ͯɻ" Attachment(image) } let hasForgottenItem = response.content.hasForgottenItem true
Evaluations framework ΞϓϦͷΠϯςϦδΣϯτػೳͷ࣭Λଌఆ͢Δ SwiftϑϨʔϜϫʔΫ • Swift Testingͱ౷߹͠ɺϓϩϯϓ τΛௐ͠ͳ͕Βਫ਼ΛଌఆͰ ͖Δ
ධՁ var dataset: ArrayLoader<ModelSample<Bool>> { ArrayLoader(samples: [ ModelSample( prompt: Prompt
{ "͜ͷը૾ʹ͍ࣸͬͯΔं྆ʹΕ͕ͳ͍͔ڭ͑ͯɻ" Attachment(imageURL: withLuggageURL) }, expected: true ), ModelSample( prompt: Prompt { "͜ͷը૾ʹ͍ࣸͬͯΔं྆ʹΕ͕ͳ͍͔ڭ͑ͯɻ" Attachment(imageURL: withoutLuggageURL) }, expected: false ), ]) }
None
Foundation Models frameworkͷҙ • σόΠεɾҬͷ੍ • AI͕ੜ͢Δग़ྗʹޡΓؚ͕·ΕΔՄೳੑ • ༻్ʹ߹ΘͤͯଞͷϑϨʔϜϫʔΫͱͷ͍͚ݕ౼
·ͱΊ • Foundation Models frameworkͷը૾ೖྗ؆୯ʹΈࠐΊΔ • Evaluations frameworkͰਫ਼ΛܧଓతʹධՁ͠ɺ࣭Λ୲อ • ಛੑΛཧղͯ͠ɺదͨ͠໘Ͱ׆༻͢Δ