Layisha imodeli yeLLM kunye netokenizer

Ukudibanisa i-DSPy kunye nee-LLM zoVavanyo lweSoftware eZizenzekelayo: IParadigm eNtsha kuQinisekiso loMgangatho

Intshayelelo

Ukudityaniswa kweMifuziselo yeeLwimi ezinkulu (LLMs) kunye neNkqubo yeSoftware eNzululwazi (i-DSPy) iye yaguqula imeko yovavanyo lwesoftware. Ngokusebenzisa amandla e-LLM kunye ne-DSPy, abaphuhlisi banokwenza inkqubo yokuvavanya ngokuzenzekelayo, ukuphucula ukuthembeka kunye nokusebenza kakuhle kubunjineli besoftware. Kweli nqaku, siza kuphonononga indlela i-DSPy enokulungelelanisa ngayo inkqubo yokuvavanya izicelo ezakhiwe kwii-LLMs, kwaye sixoxe ngeenzuzo kunye nemingeni yale paradigm evelayo.

Yintoni i-DSPy?

I-DSPy yilayibrari yePython ebonelela ngeseti yezixhobo zokuvavanya isoftware. Ivumela abaphuhlisi ukuba benze kwaye baqhube iimeko zovavanyo, ukubeka iliso kwiziphumo zovavanyo, kunye nokuhlalutya idatha yovavanyo. I-DSPy yenzelwe ukusebenza ngaphandle komthungo kunye nee-LLMs, ivumela ukuyilwa kovavanyo oluntsonkothileyo lwesakhelo sokuzisebenzela.

Zintoni iiLLM?

Ii-LLM luhlobo lwe-artificial intelligence (AI) imodeli enokuthi iqhube kwaye ivelise ulwimi olufana nomntu. Baqeqeshelwa ubuninzi bedatha yokubhaliweyo kwaye banokufunda ukuqaphela iipateni, ubudlelwane kunye nomxholo. Ii-LLM zinezicelo ezininzi kubunjineli besoftware, kubandakanya ukusetyenzwa kolwimi lwendalo, uhlalutyo lombhalo, kunye nokuveliswa kwekhowudi.

Ukudibanisa i-DSPy neeLLMs

Ukudityaniswa kwe-DSPy kunye nee-LLMs kwenza ukuba kubekho iimeko zovavanyo ezizenzekelayo ezinokuthi zenze iimeko ezinzima, ukulinganisa ukusebenzisana komsebenzisi, kunye nokuhlalutya iziphumo zovavanyo. Olu hlanganiso lubonelela ngeenzuzo ezininzi, kuquka:

Umgangatho wovavanyo ophuculweyo: I-DSPy inokuvelisa iimeko zovavanyo ezigubungela uluhlu olubanzi lweemeko, iqinisekisa ukuba isicelo sivavanywa ngokucokisekileyo.
Ukonyuka kokusebenza kakuhle: Uvavanyo oluzenzekelayo nge-DSPy kunye nee-LLMs lunciphisa ixesha kunye nomgudu ofunekayo wokuqhuba iimeko zovavanyo, ukukhulula abaphuhlisi ukuba bagxininise kweminye imisebenzi.
Ukuthembeka okuphuculweyo: I-DSPy inokubeka iliso kwiziphumo zovavanyo kwaye ilumkise abaphuhlisi kuyo nayiphi na imiba okanye iimpazamo, iqinisekisa ukuba isicelo sithembekile kwaye sizinzile.

IKhowudi yoMzekelo: Ukuziqhelanisa naMatyala oVavanyo ngeDSPy kunye neeLLMs

Nantsi ibhloko yekhowudi yomzekelo ebonisa indlela yokusebenzisa i-DSPy nge-LLM ukuzenzekelayo iimeko zovavanyo:

ngenisa dSpy
ukusuka kubaguquli bangenisa i-AutoModelForSeq2SeqLM, i-AutoTokenizer


imodeli = AutoModelForSeq2SeqLM.from_pretrained("t5-base")
uphawu = AutoTokenizer.from_pretrained("t5-base")

# Yenza ityala lovavanyo lwe-DSPy
test_case = dspy.TestCase(
    igama="Ityala loku-1",
    description="Vavanya umsebenzi wokungena",
    amanyathelo=[
        dspy.Step("Vula iphepha lokungena"),
        dspy.Step("Faka iziqinisekiso ezisebenzayo"),
        dspy.Step("Cofa iqhosha lokungena"),
    ]
)

# Yenza ityala lovavanyo usebenzisa iLLM
test_results = dspy.execute_test_case(test_case, model, tokenizer)

# Hlalutya iziphumo zovavanyo
test_results.cazulula()

Le khowudi yekhowudi ibonisa indlela yokudala i-DSPy test case, ukulayisha imodeli ye-LLM kunye ne-tokenizer, ukwenza uvavanyo lokuvavanya usebenzisa i-LLM, kwaye uhlalutye iziphumo zovavanyo.

Ephezulu 5-10 Eyeyona Isetyenziswayo Imiyalelo/iziqwengana

Nantsi eyona miyalelo/iziqwengana eziphezulu ezi-5-10 ze-DSPy neeLLMs:

dspy.TestCase(): Yenza imeko yovavanyo entsha ye-DSPy.
dspy.Step(): Yongeza inyathelo elitsha kwimeko yovavanyo yeDSPy.
dspy.execute_test_case(): Yenza uvavanyo lweDSPy usebenzisa iLLM.
imodeli = AutoModelForSeq2SeqLM.from_pretrained("t5-base"): Ilayisha imodeli yeLLM.
umqondiso = AutoTokenizer.from_pretrained("t5-base"): Ilayisha ithokeni ye-LLM.
uvavanyo_iziphumo.hlalutya(): Uhlalutya iziphumo zovavanyo lwe-DSPy.
dspy.monitor_test_results(): Ubeka iliso kwiziphumo zovavanyo lwe-DSPy.
dspy.alert_developer(): Yazisa umphuhlisi ngayo nayiphi na imiba okanye iimpazamo.
dspy.generate_test_cases(): Yenza iimeko zovavanyo ezintsha kusetyenziswa iLLM.
dspy.optimize_test_cases(): Ikhulisa ukuphunyezwa kweemeko zovavanyo usebenzisa iLLM.

Umbono

Nantsi imbonakalo ye-DSPy kunye ne-LLM udibaniso usebenzisa i-Chart.js:

{
  "type": "bar",
  "idatha": {
    "labels": ["Ityala loVavanyo 1", "Ityala loVavanyo 2", "Ityala loVavanyo 3"],
    "iiseti zedatha": [
      {
        "label": "Ixesha lokuVavanywa kwetyala",
        "idatha": [10, 20, 30],
        "backgroundColor": "rgba(255, 99, 132, 0.2)",
        "borderColor": "rgba(255, 99, 132, 1)",
        "BorderWidth": 1
      },
      {
        "label": "Inqanaba lokungaphumeleli kwetyala lovavanyo",
        "idatha": [5, 10, 15],
        "backgroundColor": "rgba(54, 162, 235, 0.2)",
        "borderColor": "rgba(54, 162, 235, 1)",
        "BorderWidth": 1
      }
    ]
  },
  "iinketho": {
    "isihloko": {
      "bonisa": yinyani,
      "umbhalo": "DSPy kunye noManyano lweLLM"
    },
    "izikali": {
      "yAxes": [
        {
          "amanqaku": {
            "beginAtZero": yinyani
          }
        }
      ]
    }
  }
}

Oku kubonwa kubonisa ixesha lokwenziwa kunye nezinga lokungaphumeleli kwamatyala amathathu ovavanyo aqhutywe kusetyenziswa i-DSPy kunye ne-LLM.

Isiphelo

Ukudityaniswa kwe-DSPy kunye nee-LLMs kuye kwayiguqula imbonakalo yesoftware yovavanyo, ivumela ukuyilwa kwamatyala ovavanyo oluzenzekelayo anokusebenzisa iimeko ezinzima, ukulinganisa ukusebenzisana komsebenzisi, kunye nokuhlalutya iziphumo zovavanyo. Ngokusebenzisa amandla e-DSPy kunye ne-LLMs, abaphuhlisi banokuhlengahlengisa inkqubo yovavanyo, bakhulise ukuthembeka kunye nokusebenza kakuhle, kunye nokuphucula umgangatho uwonke wezicelo zabo. Njengoko obu buchwephesha buqhubeka nokuvela, sinokulindela ukubona usetyenziso lwe-DSPy kunye ne-LLMs kubunjineli besoftware.

Layisha imodeli yeLLM kunye netokenizer

Have Questions?