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
LangChain Toolsの運用と改善
Search
peisuke
April 25, 2023
Technology
4
2.6k
LangChain Toolsの運用と改善
2023/4/25 LLM(GPT, PaLM等) with MLOps LT大会!!! 発表資料
peisuke
April 25, 2023
Tweet
Share
More Decks by peisuke
See All by peisuke
AI for Kids:小学生に画像認識を教えてみた話
peisuke
1
17
LangGraphで始めるマルチエージェントシステム
peisuke
13
4.3k
Self-RAG: Learning to Retrieve, Generate and Critique through Self-Reflections
peisuke
9
1.5k
Mip-NeRF 360: Unbounded Anti-Aliased Neural Radiance Fields
peisuke
0
11k
GNeRF: GAN-based Neural Radiance Field without Posed Camera
peisuke
1
770
TTS Skins: Speaker Conversion via ASR
peisuke
0
400
A Quantum Computational Approach to Correspondence Problems on Point Sets
peisuke
0
710
F0-Consistent Many-to-many Non-parallel Voice Conversion via Conditional Autoencoder
peisuke
0
190
YOLACT real-time instance segmentation
peisuke
1
290
Other Decks in Technology
See All in Technology
読んで学ぶ Amplify Gen2 / Amplify と CDK の関係を紐解く #jawsug_tokyo
tacck
PRO
1
160
AIでめっちゃ便利になったけど、結局みんなで学ぶよねっていう話
kakehashi
PRO
0
180
QA/SDETの現在と、これからの挑戦
imtnd
0
130
はじめてのSDET / My first challenge as a SDET
bun913
1
260
ブラウザのレガシー・独自機能を愛でる-Firefoxの脆弱性4選- / Browser Crash Club #1
masatokinugawa
1
480
AIエージェント開発手法と業務導入のプラクティス
ykosaka
2
1.2k
От ручной разметки к LLM: как мы создавали облако тегов в Lamoda. Анастасия Ангелова, Data Scientist, Lamoda Tech
lamodatech
0
750
勝手に!深堀り!Cloud Run worker pools / Deep dive Cloud Run worker pools
iselegant
2
380
日経電子版 for Android の技術的課題と取り組み(令和最新版)/android-20250423
nikkei_engineer_recruiting
0
400
Goの組織でバックエンドTypeScriptを採用してどうだったか / How was adopting backend TypeScript in a Golang company
kaminashi
6
6.1k
AWS全冠芸人が見た世界 ~資格取得より大切なこと~
masakiokuda
5
6.2k
AWSのマルチアカウント管理 ベストプラクティス最新版 2025 / Multi-Account management on AWS best practice 2025
ohmura
4
310
Featured
See All Featured
Scaling GitHub
holman
459
140k
Principles of Awesome APIs and How to Build Them.
keavy
126
17k
JavaScript: Past, Present, and Future - NDC Porto 2020
reverentgeek
47
5.3k
The Psychology of Web Performance [Beyond Tellerrand 2023]
tammyeverts
47
2.6k
Writing Fast Ruby
sferik
628
61k
The Success of Rails: Ensuring Growth for the Next 100 Years
eileencodes
45
7.2k
The Web Performance Landscape in 2024 [PerfNow 2024]
tammyeverts
5
520
It's Worth the Effort
3n
184
28k
Refactoring Trust on Your Teams (GOTO; Chicago 2020)
rmw
34
2.9k
Code Reviewing Like a Champion
maltzj
522
40k
Creating an realtime collaboration tool: Agile Flush - .NET Oxford
marcduiker
30
2k
Helping Users Find Their Own Way: Creating Modern Search Experiences
danielanewman
29
2.5k
Transcript
LangChain Toolsͷӡ༻ͱվળ !QFJTVLF
ࣗݾհ ໊લɿ౻ຊ ܟհ ॴଐɿ"#&+"ʢ-BCTʣ ෭ۀɿ.-Λத৺ʹ૬ஊ͔Β։ൃ·Ͱͬͯ·͢ ੋඇ͓͕͚Լ͍͞ ׆ಈɿ5XJUUFS!QFJTVLF (JUIVC IUUQTHJUIVCDPNQFJTVLF 2JJUB
IUUQTRJJUBDPNQFJTVLF 4MJEF4IBSF IUUQTXXXTMJEFTIBSFOFU'VKJNPUP,FJTVLF গ͚ͩ͠ॻ͍ͯ·͢
ࠓճͷςʔϚʹ͍ͭͯ • --.୯ಠͩͱ༻్ʹݶք͕͋Γɺ֎෦πʔϧͱͷ࿈ܞ͕େࣄ • -BOH$IBJO5PPMTͷՄೳੑ͕ੌ͍ • 5PPMΛݺͼग़ͨ͢Ίʹઃఆ͢Δ%FTDSJQUJPOΛɺ.-ͷϊϦͰ ςετɾमਖ਼Λ͢ΔΈͷఏҊͱ࣮ݧ
--.ͷݶք • --.୯ಠͰ༻ͨ͠߹ͷݶք • ֶश͍ͤͯ͞ͳ͍σʔλʹରԠͰ͖ͳ͍ͨΊɺϦΞϧλΠϜͳ࣭ ઐతɾಠࣗͷ࣭ʹ͑ΒΕͳ͍ • ౦ژͷਓޱͱ͍ͬͨಛఆͷ࣭ʹ͍ͭͯޡͬͨճΛ͕ͪ͠ • ͍͔ʹͯ͠ޡͬͨճΛग़͞ͳ͍Α͏ʹͭͭ͠ɺ༗༻ͳ݁ՌΛ
ಘΔ͔͕ࠓޙͷ׆༻ͷϙΠϯτʹͳͬͯ͘Δ
ํ • ཁ݅తͳͱ͜Ζ • --.͕ճෆՄೳͳઐతͳࣝ˞ͷΞΫηε ˞࠷৽ͷใɺ࢛ଇԋࢉɺࣾͷφϨοδɺͦͷଞ༷ʑͳΈ • ͳΔ͘ޡͬͨճΛग़͞ͳ͍Α͏ʹ͢Δ • ղܾํ
• --.୯ಠͰΘͣɺͦͷଞͷΈͱ࿈ܞ͢Δ • ݕࡧͷΈͱΈ߹Θͤͨ-MBNB*OEFYͳͲ͕͋Δ
--.ͷࠓޙͷΘΕํͷఆ ཁɺ༁ φϨοδݕࡧ ϨίϝϯυͳͲ ϑΣʔζ̍ ϑΣʔζ̎ ϑΣʔζ̏ --.୯ಠͰͷར༻ खॱͷܾ·ͬͨλ εΫ
ϧʔϧϕʔεͰ--. ͱผϩδοΫͷΈ ߹Θͤ ॊೈʹෳϩδοΫ Λݺͼग़ͯ͠ར༻ ֓ཁ ಛ ۩ମྫ ֎෦φϨοδΛ Θͳ͍ɺରͷத Ͱ݁Ͱ͖Δ ࣗ༝ͳରγες ϜɺෳࡶͳλεΫ ෳͷΈΛΈ߹Θͤ ʢ-MBNB*OEFYʣ ʢ(151MVHJOTɺ+"37*4ɺ -BOH$IBJO5PPMʣ
--.ͷࠓޙͷΘΕํͷఆ ཁɺ༁ φϨοδݕࡧ ϨίϝϯυͳͲ ϑΣʔζ̍ ϑΣʔζ̎ ϑΣʔζ̏ --.୯ಠͰͷར༻ खॱͷܾ·ͬͨλ εΫ
ϧʔϧϕʔεͰ--. ͱผϩδοΫͷΈ ߹Θͤ ॊೈʹෳϩδοΫ Λݺͼग़ͯ͠ར༻ ֓ཁ ಛ ۩ମྫ ֎෦φϨοδΛ Θͳ͍ɺରͷத Ͱ݁Ͱ͖Δ ࣗ༝ͳରγες ϜɺෳࡶͳλεΫ ෳͷΈΛΈ߹Θͤ ʢ-MBNB*OEFYʣ ʢ(151MVHJOTɺ+"37*4ɺ -BOH$IBJO5PPMTʣ ຊͷλʔήοτ
-BOH$IBJO5PPMTͷ؆୯ͳઆ໌ʢʣ • ֎෦ͷπʔϧΛॊೈʹར༻Ͱ͖ΔΈ • ಠࣗͷπʔϧΛ༻ҙ͓͖ͯ͠ʢԼਤ̍ࢀরʣɺBHFOUʹରͯ͠ ΫΤϦΛ͛ΔʢԼਤ̎ࢀরʣ • ྫɿϗςϧϨίϝϯυπʔϧ SVOϝιουͰ݁ՌΛฦ
-BOH$IBJO5PPMTͷ؆୯ͳઆ໌ʢʣ • -BOH$IBJOଆͰπʔϧΛ͏͖͔Ͳ͏͔ɺͲͷπʔϧΛ ͏͖͔Λஅͯ͠ɺπʔϧΛۦͭͭ݁͠ՌΛฦ͢ πʔϧΛ͏͔Λఆ ͏πʔϧΛܾఆ
-BOH$IBJO5PPMTͷ؆୯ͳઆ໌ʢʣ • ෳͷπʔϧΛ࿈ܞͤͨ͞ߴͳ݁ՌΛฦ͢͜ͱՄೳ ݕࡧπʔϧ ܭࢉπʔϧ
-BOH$IBJO5PPMTΛ࣮ࡍʹ͏্Ͱͷ՝ • %FTDSJQUJPOͷఆ͕ٛʮ͔ͳΓʯ͍͠ • 5PPM͕૿͑ͯ͘Δͱɺఆ௨Γʹ5PPMΛͬͯ͘Εͳ͍ • ӡ༻ͷதͰదʹ5PPMΛબͯ͘͠ΕΔΑ͏मਖ਼͍ͯ͘͠ඞཁ͕͋Δ %FTDSJQUJPO
ʮ%FTDSJQUJPO0QTʯ Λߟ͑ͯΈͨ
%FTDSJQUJPO0QTͱ • ʮ͜ͷΫΤϦʹରͯ͠ɺ͜ͷπʔϧͱ͜ͷπʔϧ͕ݺΕΔ͖ʯͱ͍͏ɺ ೖྗͱਖ਼ղͷϖΞΛ༧Ίࢁ༻ҙ͓ͯ͘͠ʢ.-ʹ͓͚Δڭࢣσʔλʣ • ࣮ࡍʹΫΤϦΛୟ͍ͯΈͯɺఆ௨Γͷπʔϧ͕ݺΕ͍ͯΔ͔Λςετ ʢ.-ʹ͓͚Δਪʣ • ݺΕ͍ͯͳ͔ͬͨΒɺ݁ՌΛ--.ʹͯ͠मਖ਼͠ɺ͜ΕΛ܁Γฦ͢ ʢ.-ʹ͓͚Δֶशʣ
͜ͷΛ͍ɺӡ༻தʹվળͷϓϩηεΛճ͢ʂ ʢͱ͍͏ϑϨʔϜϫʔΫͷఏҊʣ
%FTDSJQUJPO0QTͷ࣮ݧ • ΦεεϝͷגͷฑΛग़ྗͦ͠ͷࣄۀ༰Λग़ྗ͢ΔΈΛ ରʹ͢Δ • גͷฑ͚ͩϨίϝϯυ͞ΕͯɺͦͷϏδωεϞσϧച্ΛҰॹ ʹΓͨ͘ͳΔ • ࣗͰ৭ΜͳใΛௐͯௐࠪΛͤͣͱɺ5PPMTͷΈΛ͍ඞ ཁͳใΛҰׅͰूΊ͍ͨ
• ҎԼͷطଘπʔϧ܈Λ͍࣋ͬͯΔ͜ͱΛఆ • πʔϧ̍ɿΦεεϝͷגͷձ໊ࣾΛग़ྗ͢Δ"1* • πʔϧ̎ɿձ໊͔ࣾΒฑίʔυʹม͢Δ"1* • πʔϧ̏ɿฑίʔυʹରԠ͢Δࣄۀ༰Λग़ྗ͢Δ"1*
ςετ༻ͷ5PPMΛ࡞͓ͯ͘͠ • 5PPM͕ݺΕΔ͔͚ͩݟΕྑ͍ͷͰɺ"1*ࣗମϞοΫΞο ϓͰ0, • Սۭͷձ໊ࣾΛೖΕ͓ͯ͘͜ͱͰɺ(15ࣗମ͕͑ͳ͍Α ͏ʹ͓ͯ͘͠
%FTDSJQUJPO0QTͷϑϩʔ ਖ਼ղσʔλͷೖྗ ॳظͷ%FTDSJQUJPOઃఆ ༩͑ͨ%FTDSJQUJPOΛݩʹπʔϧΛݺͼग़͠ ਖ਼ղσʔλͱൺֱͯ͠ɺͲͷҐͷਫ਼Ͱπʔϧ͕ݺͼग़͞Ε ͔ͨΛςετ
ਫ਼্͕͕ΔΑ͏ʹ%FTDSJQUJPOΛमਖ਼ ʹΔ
%FTDSJQUJPO0QTͷϑϩʔ ਖ਼ղσʔλͷೖྗ ॳظͷ%FTDSJQUJPOઃఆ ༩͑ͨ%FTDSJQUJPOΛݩʹπʔϧΛݺͼग़͠ ਖ਼ղσʔλͱൺֱͯ͠ɺͲͷҐͷਫ਼Ͱπʔϧ͕ݺͼग़͞Ε ͔ͨΛςετ
ਫ਼্͕͕ΔΑ͏ʹ%FTDSJQUJPOΛमਖ਼ ʹΔ
ਖ਼ղσʔλͷ࡞ ΫΤϦαϯϓϧ ݺͼग़͞ΕΔ͖ 5PPM܈ͱఆग़ྗ
%FTDSJQUJPO0QTͷϑϩʔ ਖ਼ղσʔλͷೖྗ ॳظͷ%FTDSJQUJPOઃఆ ༩͑ͨ%FTDSJQUJPOΛݩʹπʔϧΛݺͼग़͠ ਖ਼ղσʔλͱൺֱͯ͠ɺͲͷҐͷਫ਼Ͱπʔϧ͕ݺͼग़͞Ε ͔ͨΛςετ
ਫ਼্͕͕ΔΑ͏ʹ%FTDSJQUJPOΛमਖ਼ ʹΔ
ॳظͷ%FTDSJQUJPOઃఆ • ͪΌΜͱֶश͞ΕΔࣄΛ֬ೝ͍ͨ͠ͷͰɺ࠷ॳదͳ༰Λ ೖΕ͓ͯ͘
%FTDSJQUJPO0QTͷϑϩʔ ਖ਼ղσʔλͷೖྗ ॳظͷ%FTDSJQUJPOઃఆ ༩͑ͨ%FTDSJQUJPOΛݩʹπʔϧΛݺͼग़͠ ਖ਼ղσʔλͱൺֱͯ͠ɺͲͷҐͷਫ਼Ͱπʔϧ͕ݺͼग़͞Ε ͔ͨΛςετ
ਫ਼্͕͕ΔΑ͏ʹ%FTDSJQUJPOΛमਖ਼ ʹΔ
मਖ਼ϓϩϯϓτͷ࡞ ݱࡏͷEFTDSJQUJPO ݺΕͨ5PPMͱݺΕͳ ͔ͬͨ5PPMΛ༩͑Δ ग़ྗϑΥʔϚοτΛࢦఆ ϢʔβͷΫΤϦࣗମʹ ༩͖͢ใ
ఆྔධՁ ਖ਼ मਖ਼ճ FQPDI • मਖ਼ʹΑͬͯਫ਼্͕͢Δ͜ͱ֬ೝ • దͳमਖ਼͕ͳ͞Εͣɺ࠷ޙʹਫ਼͕Լ͕ͬͯ͠·͏ࣄ͕͋Δ
ఆੑධՁ • ੜ͞Εͨ%FTDSJQUJPOͱΫΤϦͷՃใ %FTDSJQUJPO Ճใ
ఆੑධՁ • աֶशͯ͠͠·͏ྫʢ͜ͷล.-ͬΆ͍ʣ ˞मਖ਼༻ͷϓϩϯϓτΛ࿔͍ͬͯͨΒ࿉͞Εͨ݁Ռ ֶशσʔλʹؚ·ΕΔΫΤϦ ͱ͑Λ֮͑ͯ͠·ͬͨ
·ͱΊ • --.ͱɺͦΕҎ֎Λͭͳ͛ΔՄೳੑ͕ੌ͍ʂ • ؆୯ʹܨ͕Βͳ͍ɺ൚༻Λࢦ͢΄Ͳੑೳ্͕͕Βͳ͍ • .-ͷཁྖͰɺ%FTDSJQUJPOΛςετɾվળ͢ΔΈΛߟҊ • %FTDSJQUJPO0QTͱݴ͍ͭͭɺ·ͩ0QTͷΈ·Ͱ࡞ͬ ͍ͯͳ͍ͷͰɺࠓޙؤுΓ·͢