Layisha imodeli ye-LLM ne-tokenizer

Ukuhlanganisa i-DSPy nama-LLM Okuhlola Isofthiwe Okuzenzakalelayo: I-Paradigm Entsha Ekuqinisekiseni Ikhwalithi

Isingeniso

Ukuhlanganiswa kwamamodeli olimi olukhulu (LLMs) kanye ne-Deep Software Programming (DSPy) kuguqule isimo sokuhlola isofthiwe. Ngokusebenzisa amandla ama-LLM kanye ne-DSPy, abathuthukisi bangakwazi ukwenza inqubo yokuhlola ngokuzenzakalelayo, bathuthukise ukwethembeka nokusebenza kahle kubunjiniyela besofthiwe. Kulesi sihloko, sizohlola ukuthi i-DSPy ingakwazi kanjani ukwenza lula inqubo yokuhlola yezinhlelo zokusebenza ezakhelwe kuma-LLM, futhi sixoxe ngezinzuzo nezinselelo zalo mbono ovelayo.

Iyini i-DSPy?

I-DSPy iwumtapo wezincwadi wePython ohlinzeka ngesethi yamathuluzi okuhlola isoftware ngokuzenzakalelayo. Ivumela onjiniyela ukuthi bakhe futhi basebenzise amacala okuhlola, baqaphe imiphumela yokuhlolwa, futhi bahlaziye idatha yokuhlola. I-DSPy yakhelwe ukusebenza ngaphandle komthungo nama-LLM, ivumela ukudalwa kwezinhlaka eziyinkimbinkimbi zokuhlola ezizenzakalelayo.

Ayini ama-LLM?

Ama-LLM awuhlobo lwemodeli ye-artificial intelligence (AI) engacubungula futhi ikhiqize ulimi olufana nomuntu. Baqeqeshwe enanini elikhulu ledatha yombhalo futhi bangafunda ukubona amaphethini, ubudlelwano, nomongo. Ama-LLM anezinhlelo zokusebenza eziningi kubunjiniyela besofthiwe, okuhlanganisa ukucutshungulwa kolimi lwemvelo, ukuhlaziya umbhalo, nokukhiqizwa kwekhodi.

Ukuhlanganisa i-DSPy nama-LLMs

Ukuhlanganiswa kwe-DSPy nama-LLM kuvumela ukudalwa kwezimo zokuhlola ezizenzakalelayo ezingasebenzisa izimo eziyinkimbinkimbi, zilingise ukusebenzisana komsebenzisi, nokuhlaziya imiphumela yokuhlolwa. Lokhu kuhlanganiswa kunikeza izinzuzo ezimbalwa, okuhlanganisa:

Indlela yokuhlola ethuthukisiwe: I-DSPy ingakha amakesi okuhlola ahlanganisa inhlobonhlobo yezimo, iqinisekise ukuthi uhlelo lokusebenza luhlolisiswa kahle.
Ukusebenza kahle okuthuthukisiwe: Ukuhlola okuzenzakalelayo nge-DSPy nama-LLM kunciphisa isikhathi nomzamo odingekayo ukuze kuqhutshwe amacala okuhlola, kukhulula onjiniyela ukuze bagxile kweminye imisebenzi.
Ukuthembeka okuthuthukisiwe: I-DSPy ingaqapha imiphumela yokuhlolwa futhi yazise onjiniyela kunoma yiziphi izinkinga noma amaphutha, iqinisekise ukuthi uhlelo lokusebenza luthembekile futhi luzinzile.

Ikhodi Yesibonelo: I-Automating Test Cases nge-DSPy nama-LLMs

Nasi isibonelo sebhulokhi yekhodi ebonisa indlela yokusebenzisa i-DSPy nge-LLM ukuze wenze amacala okuhlola azenzakalelayo:

ngenisa i-dspy
kusuka kubaguquli bangenisa i-AutoModelForSeq2SeqLM, i-AutoTokenizer


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

# Dala i-DSPy test case
test_case = dspy.TestCase(
    name="Icala Lokuhlola 1",
    description="Hlola ukusebenza kokungena",
    izinyathelo=[
        dspy.Step("Vula ikhasi lokungena"),
        dspy.Step("Faka imininingwane esemthethweni"),
        dspy.Step("Chofoza inkinobho yokungena"),
    ]
)

# Yenza icala lokuhlola usebenzisa i-LLM
test_results = dspy.execute_test_case(test_case, model, tokenizer)

# Hlaziya imiphumela yokuhlolwa
test_results.hlaziya()

Leli bhulokhi lekhodi libonisa indlela yokudala ikesi lokuhlola le-DSPy, ukulayisha imodeli ye-LLM ne-tokenizer, ukhiphe icala lokuhlola usebenzisa i-LLM, futhi uhlaziye imiphumela yokuhlolwa.

Imiyalo/Amazwibela Aphezulu Okuyi-5-10

Nayi imiyalo/ amazwibela asetshenziswa kakhulu angu-5-10 we-DSPy nama-LLM:

dspy.TestCase(): Idala ikesi lokuhlola le-DSPy elisha.
dspy.Step(): Ingeza isinyathelo esisha kukesi lokuhlola le-DSPy.
dspy.execute_test_case(): Isebenzisa i-DSPy test case kusetshenziswa i-LLM.
imodeli = AutoModelForSeq2SeqLM.from_pretrained("t5-base"): Ilayisha imodeli ye-LLM.
ithokheni = AutoTokenizer.from_pretrained("t5-base"): Ilayisha ithokheni ye-LLM.
test_results.analyze(): Ihlaziya imiphumela yecala lokuhlola le-DSPy.
dspy.monitor_test_results(): Iqapha imiphumela yecala lokuhlola le-DSPy.
dspy.alert_developer(): Yazisa unjiniyela nganoma yiziphi izinkinga noma amaphutha.
dspy.generate_test_cases(): Ikhiqiza izimo zokuhlola ezintsha kusetshenziswa i-LLM.
dspy.optimize_test_cases(): Ilungiselela ukusetshenziswa kwezimo zokuhlola kusetshenziswa i-LLM.

Ukubona

Nakhu okubonwayo kokuhlanganiswa kwe-DSPy ne-LLM kusetshenziswa i-Chart.js:

{
  "type": "bar",
  "idatha": {
    "labels": ["Icala Lokuhlola 1", "Icala Lokuhlola 2", "Icala Lokuhlola 3"],
    "idathasethi": [
      {
        "label": "Isikhathi Sokuhlolwa Kwekesi",
        "idatha": [10, 20, 30],
        "backgroundColor": "rgba(255, 99, 132, 0.2)",
        "borderColor": "rgba(255, 99, 132, 1)",
        "BorderWidth": 1
      },
      {
        "label": "Izinga Lokuhluleka Kwecala Lokuhlola",
        "idatha": [5, 10, 15],
        "backgroundColor": "rgba(54, 162, 235, 0.2)",
        "borderColor": "rgba(54, 162, 235, 1)",
        "BorderWidth": 1
      }
    ]
  },
  "izinketho": {
    "isihloko": {
      "Display": iqiniso,
      "text": "DSPy kanye ne-LLM Integration"
    },
    "izikali": {
      "yAxes": [
        {
          "imikhaza": {
            "beginAtZero": kuyiqiniso
          }
        }
      ]
    }
  }
}

Lokhu kuboniswa kubonisa isikhathi sokwenza nezinga lokuhluleka kwamacala amathathu okuhlola asetshenziswe kusetshenziswa i-DSPy ne-LLM.

Isiphetho

Ukuhlanganiswa kwe-DSPy nama-LLM kuguqule ukwakheka kwesofthiwe yokuhlola, okuvumela ukudalwa kwezimo zokuhlola ezizenzakalelayo ezingasebenzisa izimo eziyinkimbinkimbi, zilingise ukusebenzisana komsebenzisi, nokuhlaziya imiphumela yokuhlolwa. Ngokusebenzisa amandla e-DSPy nama-LLM, abathuthukisi bangakwazi ukuhlela inqubo yokuhlola, bathuthukise ukwethembeka nokusebenza kahle, futhi bathuthukise ikhwalithi iyonke yezinhlelo zabo zokusebenza. Njengoba lobu buchwepheshe buqhubeka nokuvela, singalindela ukubona izinhlelo zokusebenza ezintsha ze-DSPy nama-LLM kubunjiniyela besofthiwe.

Layisha imodeli ye-LLM ne-tokenizer

Have Questions?