Mepụta RAG | B4N1

Mepụta RAG

3 months ago · b4n1 Read time: 5 min

** Ijikwa ngwa orchestration maka nhazi ngwa ngwa AI kwadoro ***

** Okwu mmalite**

Ọchịchọ na-arịwanye elu maka nhazi ngwanrọ AI na-akwado AI ebutela mmepe nke ngwá ọrụ nhazi dị iche iche na-eme ka njikọ na-enweghị isi nke Artificial Intelligence (AI) na Machine Learning (ML) n'ime usoro ngwanrọ. N'isiokwu a, anyị ga-enyocha mmejuputa ihe nhazi ngwanrọ AI na-eduzi site na iji ngwá ọrụ orchestration dị ka LangChain na DSPy, na-elekwasị anya na ijikọta RAG (Reasoning and Action Graph) na LLM (Large Language Model) ike maka nkwalite scalability, arụmọrụ, na nchekwa.

Kedu Ngwa Orchestration?

Ngwá ọrụ nhazi bụ usoro sọftụwia nke na-eme ka akpaaka nke usoro na usoro dị mgbagwoju anya site na ijikọ ọtụtụ ọrụ, ngwa na sistemụ. Ha na-enye ikpo okwu etiti maka ijikwa na ịhazi mmekọrịta dị n'etiti ihe dị iche iche, na-ahụ na arụ ọrụ nke ọma na nke nwere ike ịrụ ọrụ nke ọma.

** LangChain: Ngwá ọrụ Orchestration dabere na Python**

LangChain bụ ọba akwụkwọ Python mepere emepe nke na-enye usoro zuru oke maka iwu ụlọ sọftụwia kwadoro AI. Ọ na-enye atụmatụ dị iche iche, gụnyere:

  • ** RAG (Reasoning na Action Graph) Nchikota ***: LangChain na-enye mmejuputa RAG arụnyere, na-eme ka ịmepụta echiche dị mgbagwoju anya na eserese omume nke enwere ike iji mee ihe na-eme ka AI na-arụ ọrụ na-arụ ọrụ.
  • ** LLM (Ụdị Asụsụ buru ibu) Njikọ ***: LangChain na-akwado njikọta nke LLM, dị ka LLAMA na BERT, maka nhazi asụsụ okike na ọrụ ọgbọ.
  • ** Scalability na Performance ***: Ezubere LangChain iji jikwaa nnukwu ọrụ AI, na-enye ịrụ ọrụ nke ọma na nke nwere ike ịrụ ọrụ.

** Na-eji LangChain na-eme ihe nhazi ngwa ngwa AI na-ebugharị ***

Iji gosi mmejuputa ụkpụrụ ụlọ sọftụwia AI na-achị site na iji LangChain, ka anyị tụlee otu ihe atụ dị mfe nke chatbot nke na-eji RAG tụgharịa uche maka ajụjụ onye ọrụ wee wepụta azịza.

"Python mbubata langchain si langchain.chains bubata lm

rag = langchain.RAG( aha = "chatbot_rag", ọnụ =[ {"name": "ekele", "text": "Ndewo, kedu ka m ga-esi nyere gị aka?"}, {"name": "goodbye", "text": "Daalụ, ọ dị mma ka gị na gị kparịta ụka."}, ], ọnụ =[ {"from": "ekele", "na": "goodbye", "ọnọdụ": "user_says_goodbye"}, ], )

Mepụta LLM

llm = llm.LLM( aha = "chatbot_llm", ụdị = "llama", max_tokens=100, )

Mepụta yinye chatbot

yinye = langchain. Chain ( aha = "chatbot_chain", lm=llm, rag=akwa, )

Nwalee chatbot

user_input = "Ndewo, kedu ka ị mere?" nzaghachi = chain.run (user_input) ebipụta(nzaghachi) ``

** DSPy: Ngwa Orchestration dabere nchara**

DSPy bụ ngwá ọrụ orchestration dabeere na Rust nke na-enye arụmọrụ dị elu na nke nwere ike ime ka ọ rụọ ọrụ nhazi ngwanrọ AI kwadoro. Ọ na-enye atụmatụ dị iche iche, gụnyere:

  • ** RAG (Reasoning na Action Graph) Nchikota ***: DSPy na-enye mmejuputa RAG arụnyere, na-eme ka ịmepụta echiche dị mgbagwoju anya na eserese omume nke enwere ike iji mee ihe na-eme ka AI na-arụ ọrụ na-arụ ọrụ.
  • **LLM (Ihe Nlereanya Asụsụ buru ibu) Njikọ ***: DSPy na-akwado ntinye nke LLM, dị ka LLAMA na BERT, maka nhazi asụsụ okike na ọrụ ọgbọ.
  • ** Scalability na Performance ***: Ezubere DSPy iji jikwaa nnukwu ọrụ AI, na-enye arụmọrụ nke ọma na nke nwere ike ịrụ ọrụ.

** Iji DSPy na-eme ihe nhazi ngwa ngwa AI-chụpụrụ ***

Iji gosi mmejuputa atumatu ụlọ ọrụ sọftụwia AI na-eji DSPy, ka anyị tulee ihe atụ dị mfe nke chatbot na-eji RAG tụgharịa uche maka ajụjụ onye ọrụ wee wepụta azịza.

`` nchara jiri dspy:: yinye; jiri dspy::llm;

// Mepụta RAG ka rag = yinye::RAG:: ọhụrụ( "chatbot_rag", nke![ yinye:: Node:: ọhụrụ("ekele", "Ndewo, kedu ka m ga-esi nyere gị aka?"), yinye:: Node:: ọhụrụ("goodbye", "Daalụ, ọ dị mma ka gị na gị kparịta ụka."), ], nke![ yinye::Edge:: ọhụrụ("ekele", "goodbye", "user_ekwu_goodbye"), ], );

// Mepụta LLM ka llm = llm::LLM:: ọhụrụ( "chatbot_llm", "lama", 100, );

// Mepụta yinye chatbot ka yinye = yinye :: Chain :: ọhụrụ( "chatbot_chain", lm, akwa, );

// Nwalee chatbot let user_input = "Ndewo, kedu ka ị mere?"; ka nzaghachi = chain.run (user_input); println! ("{}", nzaghachi); ``

** Iwu 5-10 Kachasị Eji mee ihe

Nke a bụ ụfọdụ iwu na snippets ndị a na-ejikarị maka LangChain na DSPy:

LangChain: * langchain.RAG(): Na-emepụta ihe atụ RAG ọhụrụ. * langchain.LLM(): Na-emepụta ihe atụ LLM ọhụrụ. * langchain.Chain(): Na-emepụta ihe atụ yinye ọhụrụ. * rag.run(): Na-agba RAG wee wepụta nzaghachi. * llm.run(): Na-agba ọsọ LLM wee wepụta nzaghachi.

  • DSPy:
    • dspy:: yinye:: RAG:: ọhụrụ(): Na-emepụta ihe atụ RAG ọhụrụ.
    • dspy::llm::LLM:: ọhụrụ(): Na-emepụta ihe atụ LLM ọhụrụ.
    • dspy:: yinye:: Chain:: ọhụrụ(): Na-emepụta ihe atụ yinye ọhụrụ.
    • chain.run(): Na-agba agbụ ahụ wee wepụta nzaghachi.

** Nlegharị anya ***

Nke a bụ nhụta dị mfe nke ikike RAG na LLM site na iji Chart.js:

javascript const ctx = document.getElementById('myChart').getContext('2d'); const chart = Chart ọhụrụ (ctx, { ụdị: 'bar', data: { akara: ['RAG', 'LLM'], ihe ndekọ data: [{ akara: 'Ikike', data: [10, 20], ndabere Agba: [ 'rgba(255, 99, 132, 0.2)', 'rgba(54, 162, 235, 0.2)', ], Agba agba: [ 'rgba(255, 99, 132, 1)', 'rgba(54, 162, 235, 1)', ], Obosara oke: 1 }] }, nhọrọ: { akpịrịkpa: { y: { beginAtZero: eziokwu } } } }

Mkpebi

N'ikpeazụ, LangChain na DSPy bụ ngwá ọrụ nhazi siri ike nke na-eme ka mmejuputa ihe nhazi ngwanrọ AI na-eduzi site na iji ikike RAG na LLM. Site n'iji ngwá ọrụ ndị a eme ihe, ndị na-emepụta ihe nwere ike ịmepụta usoro ngwanrọ, nke na-arụ ọrụ na nke nwere nchebe nke nwere ike ijikwa ọrụ AI dị mgbagwoju anya. Site n'iji ha dị mfe na ngbanwe, LangChain na DSPy bụ nhọrọ dị mma maka iwulite ụlọ ọrụ ngwanrọ AI kwadoro.

Ndụmọdụ

Dabere na mkparịta ụka dị n'isiokwu a, ebe a bụ ụfọdụ ndụmọdụ maka ndị mmepe chọrọ iji LangChain na DSPy mejuputa nhazi ngwanrọ AI na-eduzi:

  • Malite site n'ịmụ ihe ndabere nke LangChain na DSPy, gụnyere ike RAG na LLM ha.
  • Jiri ihe atụ dị mfe nwalee, dị ka ihe atụ chatbot, ka ị nwee mmetụta maka ka ngwaọrụ ndị ahụ si arụ ọrụ.
  • Jiri nwayọọ nwayọọ gaa n'ihu n'ọrụ dị mgbagwoju anya, dị ka iwulite usoro nkwanye ma ọ bụ ọkpọkọ nhazi asụsụ eke.
  • Jiri ngalaba Iwu/Snippets nke kacha 5-10 ejiri mee ihe dị ka ntuziaka ntụaka maka iwu nkịtị na snippets.
  • Jiri Chart.js were anya gị hụ ike RAG na LLM gị iji nweta nghọta ka mma nke mmekọrịta dị n'etiti ihe dị iche iche.

Site n'ịgbaso ndụmọdụ ndị a, ndị mmepe nwere ike imeghe ikike zuru oke nke LangChain na DSPy ma wuo ụlọ ọrụ ngwanrọ AI kwadoro nke nwere ike ijikwa nnukwu ọrụ dị mgbagwoju anya ma nye ọrụ pụrụ iche na scalability.

Have Questions?

Let's Talk

Contact me on LinkedIn