Gutegura Imiyoboro Nini ya LLM Imiyoboro hamwe na DSPy na LangChain kugirango Bitezimbere DevOps | B4N1

Gutegura Imiyoboro Nini ya LLM Imiyoboro hamwe na DSPy na LangChain kugirango Bitezimbere DevOps

3 months ago · b4n1 Read time: 5 min

Gutegura Imiyoboro Nini-nini ya LLM hamwe na DSPy na LangChain kubikorwa bya DevOps

Intangiriro

Kuza kwururimi runini (LLMs) byahinduye urwego rwo gutunganya ururimi karemano (NLP) hamwe nubwenge bwa artile (AI). Ariko, uko ibintu bigoye hamwe nubunini bwa sisitemu ya LLM yiyongera, niko hakenerwa orchestre ikomeye kandi ikora neza nibikoresho byikora. Muri iki kiganiro, tuzacengera mu isi ya DSPy na LangChain, uburyo bubiri bukomeye butuma hubakwa sisitemu ya software idashobora kandi nini mugihe cya AI.

Ikibazo Cyimiyoboro Nini Nini ya LLM

Imiyoboro minini ya LLM ikubiyemo ibice byinshi, harimo gutunganya amakuru mbere, amahugurwa yicyitegererezo, hamwe nu mwanzuro. Iyi miyoboro irashobora kuba ingorabahizi, hamwe nubwinshi bwimikoranire nimikoranire hagati yibice bitandukanye. Nkuko igipimo cyiyi miyoboro cyiyongera, niko ibyago byamakosa, gutinda, hamwe nubushobozi buke.

Injira DSPy na LangChain

DSPy na LangChain nuburyo bubiri bugamije koroshya orchestre no gutangiza imiyoboro minini ya LLM. DSPy ni urwego rushingiye kuri Python rutanga uburyo bworoshye kandi bunoze bwo kubaka no gucunga imiyoboro igoye yamakuru, mugihe LangChain nurwego rushoboza kubaka sisitemu nini ya AI nini kandi ikomeye.

DSPy: Imikorere ya Python ishingiye kumiyoboro ya Data

DSPy ni Python ishingiye kumurongo utanga inzira yoroshye kandi inoze yo kubaka no gucunga imiyoboro igoye. Hamwe na DSPy, abitezimbere barashobora gukora imiyoboro yamakuru yagutse, ishobora kwihanganira, kandi yoroshye kubungabunga.

Dore urugero rwumuyoboro woroshye wubatswe na DSPy: `` python kwinjiza dspy nka dsp

data_source = dsp.Ibikoresho ( izina = "my_data_source", andika = "csv", inzira = "/ inzira / kuri / data.csv" )

Sobanura utunganya amakuru

data_processor = dsp.Umutunganya ( izina = "my_data_processor", andika = "panda", imikorere = lambda x: x.dropna () )

Sobanura ikariso

data_sink = dsp.Sink ( izina = "my_data_sink", andika = "csv", inzira = "/ inzira / kuri / ibisohoka.csv" )

Kora umuyoboro wamakuru

umuyoboro = dsp.Umuyoboro ( izina = "my_pipeline", inkomoko = [data_source], abatunganya = [data_processor], kurohama = [data_sink] )

Koresha umuyoboro

umuyoboro.run () `` Muriyi ngero, turasobanura inkomoko yamakuru, itunganya amakuru, hamwe namakuru yarohamye, hanyuma tugakora umuyoboro wamakuru uhuza ibyo bice. Umuyoboro noneho ukoreshwa ukoresheje uburyo bwa 'run ().

LangChain: Urwego rwimikorere ya AI igaragara kandi yihanganira

LangChain ni urwego rushoboza kubaka sisitemu nini kandi idashobora kwihanganira. Hamwe na LangChain, abitezimbere barashobora gukora sisitemu ya AI yoroshye kubungabunga, kugereranywa, no kwihanganira kunanirwa.

Dore urugero rwa sisitemu yoroshye ya AI yubatswe na LangChain: `` python gutumiza mu mahanga nka lc

Sobanura icyitegererezo

icyitegererezo = lc.Model ( izina = "my_model", andika = "transformateur", inzira = "/ inzira / kuri / icyitegererezo" )

Sobanura ikibazo

ikibazo = lc.Ibisobanuro ( izina = "my_prompt", inyandiko = "Ubu ni urugero rw'icyitegererezo" )

Kora sisitemu ya AI

ai_system = lc.Uburyo ( izina = "my_ai_system", icyitegererezo = icyitegererezo, ikibazo = ikibazo )

Koresha sisitemu ya AI

ai_system.run () `` Mururugero, turasobanura icyitegererezo, ikibazo, hanyuma tugakora sisitemu ya AI ihuza ibyo bice. Sisitemu ya AI noneho ikoreshwa ikoresheje uburyo bwa 'run ().

Gutegura Imiyoboro Nini ya LLM hamwe na DSPy na LangChain

Kugirango utegure imiyoboro minini ya LLM hamwe na DSPy na LangChain, dushobora gukoresha uburyo bukurikira:

  1. Sobanura umuyoboro wamakuru ukoresheje DSPy ihuza amasoko menshi yamakuru, abatunganya, hamwe na sink.
  2. Sobanura sisitemu ya AI ukoresheje LangChain ihuza moderi nyinshi, ibisobanuro, nibisohoka.
  3. Koresha DSPy gucunga imiyoboro yamakuru na LangChain kugirango ucunge sisitemu ya AI.
  4. Koresha gahunda nka Apache Airflow cyangwa Kubernetes kugirango utegure ishyirwa mubikorwa ryumuyoboro wamakuru na sisitemu ya AI.

Dore urugero rwimyandikire yoroshye ikoresha DSPy na LangChain: `` python kwinjiza dspy nka dsp gutumiza mu mahanga nka lc gutumiza apache_ibikorwa nka af

Sobanura umuyoboro wamakuru

amakuru_umuyoboro = dsp.Umuyoboro ( izina = "my_data_pipeline", inkomoko = [dsp. Ibikoresho (izina = "my_data_source", andika = "csv", inzira = "/ inzira / kuri / data.csv")], gutunganya = [dsp. kurohama = [dsp.Sink (izina = "my_data_sink", andika = "csv", inzira = "/ inzira / kuri / ibisohoka.csv")] )

Sobanura sisitemu ya AI

ai_system = lc.Uburyo ( izina = "my_ai_system", icyitegererezo = lc.Model (izina = "my_model", andika = "transformateur", inzira = "/ inzira / kuri / icyitegererezo"), ikibazo = lc.Ibisobanuro (izina = "my_prompt", inyandiko = "Ubu ni urugero rw'icyitegererezo") )

Teganya ishyirwa mubikorwa ryumuyoboro wamakuru na sisitemu ya AI

af.urutonde_ibibazo ( umurimo_id = "my_task", dag_id = "my_dag", umuyoboro = amakuru_umuyoboro, sisitemu = ai_system ) `` ` Muri uru rugero, turasobanura umuyoboro wamakuru hamwe na sisitemu ya AI dukoresheje DSPy na LangChain, hanyuma tugashyiraho gahunda yo gukora ibyo bice dukoresheje Apache Airflow.

Hejuru 5-10 Amabwiriza Yakoreshejwe Byinshi / Udukoryo

Dore top 5-10 zikoreshwa cyane / ibisobanuro kuri DSPy na LangChain:

  1. dsp. Umuyoboro (izina =" my_pipeline ", inkomoko = [isoko], abatunganya = [umutunganya], kurohama = [kurohama))
  2. lc.Model (izina =" my_model ", andika =" transformateur ", inzira =" / inzira / kuri / icyitegererezo ")
  3. `Lc.Prompt (izina =" my_prompt ", inyandiko =" Ubu ni icyitegererezo ")
  4. dsp. Ibikoresho (izina =" my_data_source ", andika =" csv ", inzira =" / inzira / kuri / data.csv ")
  5. dsp.Umutunganya (izina =" my_data_processor ", andika =" panda ", imikorere = lambda x: x.dropna ())
  6. dsp.Sink (izina =" my_data_sink ", andika =" csv ", inzira =" / inzira / kuri / ibisohoka.csv ")
  7. `lc.Uburyo (izina =" my_ai_system ", icyitegererezo = icyitegererezo, ikibazo = ikibazo)
  8. af.schedule_task (task_id =" my_task ", dag_id =" my_dag ", umuyoboro = umuyoboro, sisitemu = sisitemu)
  9. dsp.run_pipeline (umuyoboro)
  10. Lc.run_system (sisitemu)

Kubona amashusho

Hano hari amashusho yuburyo bwa DSPy na LangChain ukoresheje Chart.js: json { "ubwoko": "akabari", "amakuru": { "ibirango": ["DSPy", "LangChain"], "imibare": [ { "label": "Ibiranga", "amakuru": [10, 15], "backgroundColor": [ "rgba (255, 99, 132, 0.2)", "rgba (54, 162, 235, 0.2)" ], "umupaka w'amabara": [ "rgba (255, 99, 132, 1)", "rgba (54, 162, 235, 1)" ], "Ubugari bwumupaka": 1 } ] }, "amahitamo": {} } ` Iyi visualisation yerekana umubare wibintu muri DSPy na LangChain.

Umwanzuro

Mu gusoza, DSPy na LangChain nuburyo bubiri bukomeye butuma hubakwa sisitemu ya software idashobora kandi nini mugihe cya AI. Ukoresheje iyi mikorere, abitezimbere barashobora gukora imiyoboro minini ya LLM yoroshye kubungabunga, kugereranywa, no kwihanganira kunanirwa. Hifashishijwe orchestre hamwe nibikoresho byikora nka Apache Airflow na Kubernetes, abitezimbere barashobora guteganya ishyirwa mubikorwa ryiyi miyoboro na sisitemu, byoroshye gucunga sisitemu ya AI igoye.

Akazi kazoza

Akazi kazoza karimo:

  1. Gutezimbere ibintu byinshi kuri DSPy na LangChain.
  2. Kunoza imikorere nubunini bwa DSPy na LangChain.
  3. Gutezimbere ibikoresho byinshi bya orchestre nibikoresho byikora kuri DSPy na LangChain.
  4. Gukora amashusho menshi hamwe nibibaho bya DSPy na LangChain.

Reba

  1. Inyandiko ya DSPy: https://dspy.readthedocs.io/en/latest/
  2. Inyandiko ya LangChain: https://langchain.readthedocs.io/en/latest/
  3. Apache Airflow ibyangombwa: https://airflow.apache.org/docs/apache-airflow/stable/index.html
  4. Kubernetes ibyangombwa: https://kubernetes.io/docs/urugo/

Have Questions?

Let's Talk

Contact me on LinkedIn