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
RAGの精度が全然上がらない!! AOSSを使った社内RAG開発の反省
Search
とすり
December 13, 2024
2
200
RAGの精度が全然上がらない!! AOSSを使った社内RAG開発の反省
とすり
December 13, 2024
Tweet
Share
More Decks by とすり
See All by とすり
GraphRAGの仕組みまるわかり
tosuri13
9
550
NL2SQLを活用したExcelの生成AI利用アプローチ
tosuri13
0
47
AWS Chaliceで始める爆速サーバレスチャットボット開発!!
tosuri13
1
210
Amazon BedrockでサーバレスなAIお料理ボットを作成する!!
tosuri13
3
600
React + TextAliveでカッコいいLyric Applicatioinを作ろう!!
tosuri13
1
690
Radix UI & shadcn/uiのススメ
tosuri13
0
140
Amazon BedrockとOpenSearch Serviceでなんでも答えられる社内RAGを作成する!!
tosuri13
4
690
Featured
See All Featured
Building an army of robots
kneath
306
45k
ReactJS: Keep Simple. Everything can be a component!
pedronauck
667
120k
Testing 201, or: Great Expectations
jmmastey
43
7.6k
A Tale of Four Properties
chriscoyier
160
23k
Done Done
chrislema
184
16k
The Language of Interfaces
destraynor
158
25k
Designing for Performance
lara
610
69k
A Modern Web Designer's Workflow
chriscoyier
695
190k
Mobile First: as difficult as doing things right
swwweet
223
9.7k
"I'm Feeling Lucky" - Building Great Search Experiences for Today's Users (#IAC19)
danielanewman
229
22k
Building Better People: How to give real-time feedback that sticks.
wjessup
367
19k
Sharpening the Axe: The Primacy of Toolmaking
bcantrill
44
2.4k
Transcript
RAGͷਫ਼͕શવ্͕Βͳ͍!! AOSSΛͬͨࣾRAG։ൃͷল 2024.12.13 JAWS-UG ਆށ #3 ΕେLTେձ @tosuri13
ͱ͢Γ @tosuri13 MOTEXגࣜձࣾ ࡶ༻ܥΤϯδχΞ(ࣗশ) Stor a ge Browser for Am
a zon S3Λ Amplifyൈ͖Ͱ͑ͳ͍͔ࡧதͰ͢🥺
ࣾจॻΛѻ͑ΔRAGγεςϜΛAWSͰ։ൃɾӡ༻த… RAGͷਫ਼্ʹେۤઓ!! ՝লϙΠϯτʹ͍͍ͭͯͨ͠ͱࢥ͍·͢ ↑ ࠓ7݄ͷBedrock Night in େࡕͰͨ͠RAGγεςϜͰ͢☺
ͬ͘͟Γݱঢ়ͷAWSߏਤ
RAGͷਫ਼͕શવ্͕Βͳ͍!!
AOSSʹυΩϡϝϯτΛେྔೖͯ͠ӡ༻։࢝!! → ͔͠͠ظ͢Δճ͕શવฦͬͯ͜ͳ͍!! ݪҼΛ୳Δ͜ͱʹ… ͜ͷػೳ͕Ճ͞Εͨͷͬͯ Ͳͷόʔδϣϯ͔Β? ͳͳ Βͳ͍Ͱ͢ ࣾRAGϘοτ
RAGʹ͓͍ͯҰ൪ॏཁͳϑΣʔζͲ͔͜? https:// a ws. a m a zon.com/jp/blogs/news/ a -pr
a ctic a l-guide-to-improve-r a g-systems-with- a dv a nced-r a g-on- a ws ΑΓҾ༻
https:// a ws. a m a zon.com/jp/blogs/news/ a -pr a
ctic a l-guide-to-improve-r a g-systems-with- a dv a nced-r a g-on- a ws ΑΓҾ༻ ҰൠతʹRetrieveॲཧͩͱݴΘΕ͍ͯΔ (ແؔͳσʔλΛͯ͠͠·͏ͱɺͲΕ͚ͩᘳͳLLMͰదͳճ͕Ͱ͖ͳ͍ͨΊ)
ͦͷͨΊɺRetrieveॲཧΛ࠷దԽ͢ΔΑ͏ʹ ༷ʑͳΞϓϩʔνͰ࣮ݧɺௐࠪΛॏͶΔ… Am a zon BedrockͷຒΊࠐΈϞσϧΛม͑ͯΈͨΓ… (EmbeddingҎ֎ʹϝλσʔλΛೖΕͨΓ) AOSSͷANNϥΠϒϥϦΛม͑ͨΓ HNSWͷύϥϝʔλΛ͍͔͍ͭͬͯ͘͡ΈͨΓ… F
a iss͔nmslib͔…
ͦͷͨΊɺRetrieveॲཧΛ࠷దԽ͢ΔΑ͏ʹ ༷ʑͳΞϓϩʔνͰ࣮ݧɺௐࠪΛॏͶΔ… Am a zon BedrockͷຒΊࠐΈϞσϧΛม͑ͯΈͨΓ… (EmbeddingҎ֎ʹϝλσʔλΛೖΕͨΓ) AOSSͷANNϥΠϒϥϦΛม͑ͨΓ HNSWͷύϥϝʔλΛ͍͔͍ͭͬͯ͘͡ΈͨΓ… F
a iss͔nmslib͔… …͕!! ճશવվળ͞Εͳ͍!!
ճͷ࣭ԼΛট͍͍ͯͨຊͷݪҼ… https:// a ws. a m a zon.com/jp/blogs/news/ a -pr
a ctic a l-guide-to-improve-r a g-systems-with- a dv a nced-r a g-on- a ws ΑΓҾ༻
ͩͬͨ͜͜!! https:// a ws. a m a zon.com/jp/blogs/news/ a -pr
a ctic a l-guide-to-improve-r a g-systems-with- a dv a nced-r a g-on- a ws ΑΓҾ༻
ݪҼௐࠪΛਐΊΔ্Ͱ͔ͬͨͷ… ʮͦͦͷυΩϡϝϯτ͕͓͔͍͠!!ʯͱ͍͏͜ͱ ֦ுࢠͱ࣮ࡍͷத͕ Կނ͔ҧ͏ϑΝΠϧ ΫΫΫ… ↑ mdʹٖଶ͢Δxml ωਃExcelͷଘࡏ (ҳ͍ͨ͠ํͷExcelγʔτ) ແҙຯͳใͰ
ຒΊਚ͘͞ΕͨHTML ϦϯΫେྔʹ࣋ͬͯ·͢
ݪҼௐࠪΛਐΊΔ্Ͱ͔ͬͨͷ… ʮͦͦͷυΩϡϝϯτ͕͓͔͍͠!!ʯͱ͍͏͜ͱ ֦ுࢠͱ࣮ࡍͷத͕ Կނ͔ҧ͏ϑΝΠϧ ΫΫΫ… ↑ mdʹٖଶ͢Δxml ωਃExcelͷଘࡏ (ҳ͍ͨ͠ํͷExcelγʔτ) ແҙຯͳใͰ
ຒΊਚ͘͞ΕͨHTML ϦϯΫେྔʹ࣋ͬͯ·͢ ݕࡧͷϊΠζʹͳΓɺదͳσʔλΛ ఏڙͰ͖ͳ͘ͳ͍ͬͯΔ!!
લॲཧ + νϟϯΩϯάΛ͘ݟ͍ͯͨͷ͕ݪҼ!! (L a ngCh a inඋ͚͑ͷϩʔμʔʹॲཧΛؙ͍ͤͯͨ͠ѱ͔ͬͨ…) ݕূ࣌ ࣮ӡ༻࣌
͋Δఔ៉ྷͳυΩϡϝϯτΛ ͬͯݕূͨͨ͠Ίʹؾ͚ͣ… શવେৎͩͳ!! ࣮ࡍͷυΩϡϝϯτۄੴࠞަঢ়ଶ!! اۀͷ࣮ଶʹԊͬͨॲཧΛΉඞཁ͕͋Δ
֤υΩϡϝϯτͱਅʹ͖߹͍ ͦΕͧΕ࠷దͳܗͰม͍ͯ͘͠ॲཧΛߦͳͬͨ ࣄલʹυΩϡϝϯτΛ֬ೝ͠ ѻ͑ͳ͍ͷೖΕͳ͍ தXML͡ΌΜ!! ωਃ͔Β͑Δ෦Λநग़ LLMͰཁͯ͠Ϩίʔυܗࣜʹ BS4Ͱղੳͯ͠ ෆཁͳλάΛΫϦʔχϯά
͢ΔͱɺಛʹߏઃఆมΘ͍ͬͯͳ͍ͷʹ ظ͢Δճ͕͑ΔΑ͏ʹͳͬͨ!! ͜ͷػೳ͕Ճ͞Εͨͷͬͯ Ͳͷόʔδϣϯ͔Β? όʔδϣϯX.X͔ΒͰ͢ ࣾRAGϘοτ ͍͍ײ͡ͷػೳ͕૿͑·ͨ͠
None
·ͱΊ
ɾAIٕज़ཁૉʹਅʹ͖߹͓͏!! ɾRAGΛݕ౼͢ΔલʹυΩϡϝϯτཧΛ!! → ͦͦRAGΛΘͣͱɺ͙͢ʹυΩϡϝϯτΛݟ͚ͭΒΕΔঢ়ଶ͕·͍͠Ͱ͢ → ීஈ͔Β៉ྷͳυΩϡϝϯτΛॻ͖·͠ΐ͏!! → લॲཧΛͤͣʹదʹಥͬࠐΉͱμϝͩͱ͍͏ͷ͕Α͔͘Γ·ͨ͠ → طଘͷAIαʔϏεͤͰͳ͘ɺటष͍͍ͯ͘ͷͰਅʹ͖߹͏͜ͱ͕େࣄͰͨ͠
Th a nk you for listening!! @tosuri13 ← Α͔ͬͨΒTwitterϑΥϩʔͯ͠Ͷ