Guhuza RAG hamwe na LLMs ya Dynamic Software Architecture: Ubuyobozi bufatika

Intangiriro

Mu myaka yashize, urwego rwububiko bwa software rwagize impinduka zikomeye hamwe no kuza kwa Moderi nini y'ururimi (LLMs) hamwe na tekinoroji ya Retrieval-Augmented Generation (RAG). Imikoranire hagati yikoranabuhanga ryombi ifite ubushobozi bwo guhindura uburyo dushushanya no guteza imbere sisitemu ya software. Muri iyi ngingo, tuzasesengura guhuza RAG na LLMs kubwububiko bwa software ikora, harimo ingamba zo gushyira mubikorwa ukoresheje ibikoresho bya orchestre nka LangChain na DSPy.

RAG ni iki?

RAG nubuhanga bukomatanya imbaraga za Retrieval-ishingiye hamwe na Generation-ishingiye kubisohokayandikiro byujuje ubuziranenge. Igitekerezo cyibanze inyuma ya RAG nugukoresha uburyo bwo kugarura ibintu kugirango uhitemo inyandiko cyangwa ibice bijyanye na corpus nini, hanyuma ukoreshe icyitegererezo cyibisekuru kugirango ubyare inyandiko ishingiye kuri ibi bice byakuwe. Ubu buryo bwerekanye iterambere ryinshi mubikorwa nko gusubiza ibibazo, incamake yinyandiko, hamwe na AI yo kuganira.

LLM ni iki?

LLMs ni ubwoko bwubwenge bwubukorikori butozwa kubwinshi bwamakuru yinyandiko kugirango habeho ibisubizo bisa nururimi rwabantu. Izi moderi mubusanzwe zishingiye kubikorwa bya transformateur kandi byageze kubisubizo bigezweho muburyo butandukanye bwo gutunganya ururimi karemano. LLMs irahuze cyane kandi irashobora gukoreshwa mubikorwa nko guhindura ururimi, incamake yinyandiko, hamwe na AI yo kuganira.

Kuki Guhuza RAG na LLMs?

Kwishyira hamwe kwa RAG hamwe na LLMs bitanga inyungu nyinshi muburyo bwa software ikora:

2. ** Kongera imbaraga **: RAG irashobora kugabanya kubara hejuru ya LLMs uhitamo gusa ibyangombwa cyangwa ibice byingenzi, biganisha kubisubizo byihuse.
3. ** Byongerewe imbaraga **

Ingamba zo Gushyira mu bikorwa

Hariho ingamba nyinshi zo gushyira mubikorwa guhuza RAG na LLMs, harimo:

1. Ifasha intera nini ya LLMs kandi irashobora gukoreshwa mugushira mubikorwa RAG yubatswe.
2. ** DSPy **: DSPy ni isomero rya Python ritanga uburyo bworoshye kandi bunoze bwo gushyira mubikorwa RAG ishingiye. Ifasha intera nini ya LLMs kandi irashobora gukoreshwa mugushira mubikorwa RAG yubatswe.

Urugero rwa Kode

Hano hari code yoroshye yerekana uburyo bwo gukoresha LangChain kugirango uhuze RAG na LLMs: `` python gutumiza mu mahanga Kuva kumurongo.iminyururu itumiza RAGChain Kuva kumurongo.llms itumiza LLaMA

llama = LLaMA ()

Kora icyitegererezo cya RAGChain

rag_chain = RAGChain (llama, kugarura_model = "t5-nto")

Sobanura ikibazo

prompt = "Umurwa mukuru w'Ubufaransa ni uwuhe?"

Kora ibisohoka

ibisohoka = rag_chain.run (ikibazo)

icapiro (ibisohoka) `` ` Uru rugero rwimyandikire yerekana uburyo wakoresha LangChain kugirango ukore moderi ya RAGChain ihuza moderi ya LLaMA hamwe nuburyo bwo kugarura ibintu. Moderi ya RAGChain noneho ikoreshwa mugutanga inyandiko zisohoka zishingiye kukibazo.

Hejuru 5-10 Amabwiriza Yakoreshejwe Byinshi / Udukoryo

Hano haribintu 5-10 byambere bikoreshwa cyane / ibisobanuro byo guhuza RAG na LLMs:

1. `langchain.
2. Llama.run (prompt): Yibyara inyandiko isohoka ishingiye kukibazo ukoresheje moderi ya LLaMA.
3. rag_chain.run (prompt): Yibyara inyandiko isohoka ishingiye kukibazo ukoresheje icyitegererezo cya RAGChain.
4. `langchain. 5 ..
5. rag_chain.get_output (prompt): Subiza inyandiko yasohotse kuri moderi ya RAGChain.
6. `langchain.
7. Llama.get_ibisohoka (ikibazo): Subiza inyandiko yasohotse kuri moderi ya LLaMA.
8. rag_chain.get_input (prompt): Subiza inyandiko yinjiza kuri moderi ya RAGChain.
9. `langchain.

Kubona amashusho

Hano haribintu byoroshye byerekana moderi ya RAGChain ukoresheje Chart.js: json { "ubwoko": "akabari", "amakuru": { "ibirango": ["LLaMA", "RAGChain"], "imibare": [ { "label": "Ukuri", "amakuru": [0.8, 0.9], "backgroundColor": ["# FF6384", "# 36A2EB"] } ] }, "amahitamo": { "umutwe": { "kwerekana": ukuri, "inyandiko": "Icyitegererezo cya RAGChain" }, "umunzani": { "yAxes": [ { "amatiku": { "gutangiraAtZero": ukuri } } ] } } } ` Iyi visualisation yerekana ukuri kwa LLaMA nicyitegererezo cya RAGChain.

Umwanzuro

kwishyira hamwe kwa RAG hamwe na LLMs bitanga inyungu nyinshi muburyo bwububiko bwa software ikora, harimo kunonosora neza, kongera imikorere, no kongera ubworoherane. LangChain na DSPy nibikoresho bibiri bizwi cyane bya orchestre bishobora gukoreshwa mugushira mubikorwa RAG ishingiye. Mugukurikiza ingamba zo gushyira mubikorwa hamwe nurugero rwimyandikire yatanzwe muriyi ngingo, abashinzwe iterambere barashobora gukora imiterere ya RAG yubatswe ihuza LLMs hamwe nuburyo bwo kugarura ibintu. Hejuru 5-10 ikoreshwa cyane amategeko / uduce two guhuza RAG na LLMs nayo yatanzwe kugirango yerekanwe.