Na-ahazi pipeline LLM buru ibu na DSPy na LangChain maka DevOps emelitere | B4N1

Na-ahazi pipeline LLM buru ibu na DSPy na LangChain maka DevOps emelitere

6 months ago · b4n1 Read time: 5 min

** Nhazi pipeline LLM buru ibu na DSPy na LangChain maka DevOps emelitere ***

** Okwu mmalite**

Ọbịbịa nke Ụdị Asụsụ buru ibu (LLM) agbanweela ngalaba nke Nhazi Asụsụ eke (NLP) na ọgụgụ isi Artificial (AI). Otú ọ dị, ka mgbagwoju anya na ọnụ ọgụgụ nke sistemu dabeere na LLM na-abawanye, otú ahụ ka ọ dị mkpa maka ngwa ngwa na ngwa ngwa siri ike ma dị mma. N'isiokwu a, anyị ga-abanye n'ime ụwa nke DSPy na LangChain, usoro abụọ dị ike nke na-eme ka ịmepụta usoro ngwanrọ na-agbanwe agbanwe na nke nwere ike ime na afọ AI.

** Ihe ịma aka nke pipeline LLM buru ibu ***

Pipeline LLM buru ibu gụnyere ọtụtụ ihe, gụnyere nhazi data, ọzụzụ ụdị, na ntinye. Pipeline ndị a nwere ike ịdị mgbagwoju anya, yana ọtụtụ ịdabere na mmekọrịta dị n'etiti ihe dị iche iche. Ka ọnụ ọgụgụ nke pipeline ndị a na-abawanye, otu ahụ ka ihe ize ndụ nke njehie, igbu oge na ihe mgbochi akụ na-abawanye.

** Tinye DSPy na LangChain ***

DSPy na LangChain bụ usoro abụọ na-achọ ime ka ọ dị mfe nhazi na akpaaka nke nnukwu pipeline LLM. DSPy bụ usoro nke dabeere na Eke Ọgba nke na-enye ụzọ dị mfe ma dị irè iji wuo ma jikwaa pipeline data mgbagwoju anya, ebe LangChain bụ usoro nke na-enyere aka iwulite usoro AI nke na-agbanwe agbanwe na nke na-agbanwe agbanwe.

** DSPy: Usoro dabere na Python maka pipeline data ***

DSPy bụ usoro sitere na Python nke na-enye ụzọ dị mfe ma dị mma iji wuo na jikwaa pipeline data dị mgbagwoju anya. Site na DSPy, ndị mmepe nwere ike ịmepụta pipeline data nke nwere ike ịgbatị, na-agbanwe agbanwe ma dị mfe idowe.

Nke a bụ ọmụmaatụ nke pipeline data dị mfe nke ejiri DSPy rụọ: "Python mbubata dspy ka dsp

data_source = dsp.Isi iyi( aha = "my_data_source", ụdị = "csv", ụzọ = "/ụzọ/to/data.csv" )

Kọwaa onye nrụpụta data

data_processor = dsp.Processor( aha = "my_data_processor", ụdị = "pandas", func=lambda x: x.dropna() )

Kọwaa sink data

data_sink = dsp.Sink( aha = "my_data_sink", ụdị = "csv", ụzọ = "/ụzọ/to/output.csv" )

Mepụta pipeline data

pipeline = dsp. aha = "my_pipeline", isi mmalite=[data_source], nhazi = [data_processor], sinks=[data_sink] )

Gbaa pipeline

pipeline.run () `` N'ihe atụ a, anyị na-akọwapụta isi iyi data, nhazi data, na sink data, wee mepụta pipeline data nke jikọtara ihe ndị a. A na-eji usoro 'run()' na-agba ọkpọkọ a.

** LangChain: Usoro maka Sistemụ AI nwere ike ịgbatị na nke na-agbanwe agbanwe ***

LangChain bụ usoro nke na-enyere aka iwulite sistemu AI nke nwere ike ịgbatị ma na-agbanwe agbanwe. Site na LangChain, ndị mmepe nwere ike ịmepụta usoro AI nke dị mfe idobe, scalable, na resilience ọdịda.

Nke a bụ ọmụmaatụ nke sistemụ AI dị mfe nke ejiri LangChain wuo: "Python mbubata langchain dị ka lc

Kọwaa ihe nlereanya

nlereanya = lc. Model ( aha = "my_model", ụdị = "transformer", ụzọ = "/ ụzọ / gaa / ihe nlereanya" )

Kọwaa ngwa ngwa

ozugbo = lc. aha = "my_prompt", text="Nke a bụ ihe nlere anya" )

Mepụta sistemụ AI

ai_system = lc.AISystem( aha = "my_ai_system", ụdị = ihe nlereanya, ngwa ngwa = ngwa ngwa )

Gbaa usoro AI

ai_system.run() `` N'ihe atụ a, anyị na-akọwapụta ihe nlereanya, ngwa ngwa, wee mepụta usoro AI nke jikọtara ihe ndị a. A na-eji usoro 'run()' arụ usoro AI.

** Nhazi pipeline LLM buru ibu nwere DSPy na LangChain ***

Iji hazie pipeline LLM buru ibu na DSPy na LangChain, anyị nwere ike iji usoro a:

  1. Kọwaa pipeline data site na iji DSPy nke na-ejikọta ọtụtụ isi mmalite data, nhazi, na sinks.
  2. Kọwaa usoro AI site na iji LangChain nke na-ejikọta ọtụtụ ụdị, mkpali, na mmepụta.
  3. Jiri DSPy jikwaa pipeline data na LangChain iji jikwaa usoro AI.
  4. Jiri onye nhazi oge dị ka Apache Airflow ma ọ bụ Kubernetes na-eme ndokwa maka ogbugbu nke pipeline data na usoro AI.

Nke a bụ ọmụmaatụ nke edemede egwu egwu dị mfe nke na-eji DSPy na LangChain: "Python mbubata dspy ka dsp mbubata langchain dị ka lc mbubata apache_airflow dị ka af

Kọwaa pipeline data

data_pipeline = dsp.Pipeline( aha = "my_data_pipeline", Source=[dsp.Source (aha = "my_data_source", ụdị = "csv", ụzọ = "/ụzọ/to/data.csv")], nhazi = [dsp.Processor (aha = "my_data_processor", ụdị = "pandas", func = lambda x: x.dropna ())], sinks = [dsp.Sink (aha = "my_data_sink", ụdị = "csv", ụzọ = "/ụzọ/to/output.csv")] )

Kọwaa sistemụ AI

ai_system = lc.AISystem( aha = "my_ai_system", ụdị = lc.Model (aha = "my_model", ụdị = "transformer", ụzọ = "/ ụzọ / na / ihe nlereanya"), Quick=lc.Prompt(aha = "my_prompt", ederede = "Nke a bụ ihe nlere anya") )

Hazie ogbugbu nke pipeline data na sistemụ AI

af.schedule_task( task_id = "My_ọrụ", dag_id = "my_dag", pipeline = data_pipeline, sistemu=ai_sistemu ) `` N'ihe atụ a, anyị na-akọwa pipeline data na usoro AI site na iji DSPy na LangChain, wee hazie nhazi nke ihe ndị a site na iji Apache Airflow.

** Iwu 5-10 Kachasị Eji mee ihe

Nke a bụ iwu/snippets 5-10 kacha eji eme ihe maka DSPy na LangChain:

  1. dsp.Pipeline (aha = "my_pipeline", isi mmalite = [isi iyi], processors = [processor], sinks = [sink])
  2. lc.Model (aha = "my_model", ụdị = "transformer", ụzọ = "/ ụzọ / na / ihe nlereanya")
  3. lc.Prompt(aha = "my_prompt", ederede = "Nke a bụ ihe nlele ozugbo")
  4. dsp.Source (aha = "my_data_source", ụdị = "csv", ụzọ = "/ụzọ/to/data.csv")
  5. dsp.Processor(aha = "my_data_processor", ụdị = "pandas", func=lambda x: x.dropna())
  6. dsp.Sink (aha = "my_data_sink", ụdị = "csv", ụzọ = "/ụzọ/to/output.csv")
  7. lc.AISystem(aha = "my_ai_system", model=model, ozugbo = ngwa ngwa)
  8. af.schedule_task (ọrụ_id = "my_task", dag_id = "my_dag", pipeline = pipeline, usoro = sistem)
  9. dsp.run_pipeline(ọkpọkpọ)
  10. lc.run_system(sistemu)

** Nlegharị anya ***

Nke a bụ nhụta nke usoro DSPy na LangChain na-eji Chart.js: Json { "ụdị": "ogwe", "data": { "labels": ["DSPy", "LangChain"], "datasets": [ { "label": "atụmatụ", "data": [10, 15], "backgroundAgba": [ "rgba(255, 99, 132, 0.2)", "rgba (54, 162, 235, 0.2)" ], "BorderAgba": [ "rgba (255, 99, 132, 1)", "rgba (54, 162, 235, 1)" ], "BorderWiidth": 1 } ] }, "Nhọrọ": {} } `` Ọhụụ a na-egosi ọnụọgụ njirimara dị na DSPy na LangChain.

Mkpebi

DSPy na LangChain bụ usoro abụọ dị ike nke na-enyere aka iwulite usoro ngwanrọ na-agbanwe agbanwe na nke nwere ike ime na afọ AI. Site n'iji usoro ndị a, ndị mmepe nwere ike ịmepụta nnukwu ọkpọkọ LLM nke dị mfe idowe, na-agbanwe agbanwe, na nkwụsị nke ọdịda. Site n'enyemaka nke orchestration na ngwá ọrụ akpaaka dị ka Apache Airflow na Kubernetes, ndị mmepe nwere ike ịhazi nhazi nke pipeline na usoro ndị a, na-eme ka ọ dị mfe ijikwa usoro AI mgbagwoju anya.

** Ọrụ n'ọdịnihu ***

Ọrụ n'ọdịnihu gụnyere:

  1. Ịmepụta atụmatụ ndị ọzọ maka DSPy na LangChain.
  2. Ịkwalite arụmọrụ na scalability nke DSPy na LangChain.
  3. Ịmepụta ihe ndị ọzọ orchestration na akpaaka maka DSPy na LangChain.
  4. Ịmepụta ihe ọhụụ na dashboard maka DSPy na LangChain.

** Ntụaka ***

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

Have Questions?

Let's Talk

Contact me on LinkedIn