Laola mohlala oa LLM le tokenizer

Ho kopanya DSPy le LLM bakeng sa Teko ea Automated Software: Paradigm e Ncha ho Tiisetso ea Boleng

Selelekela

Ho kopanngoa ha Meetso ea Lipuo tse Khōlō (LLMs) le Deep Software Programming (DSPy) ho fetotse mokhoa oa ho hlahloba software. Ka ho matlafatsa bokhoni ba LLMs le DSPy, bahlahisi ba ka iketsetsa ts'ebetso ea tlhahlobo, ba matlafatsa ts'epo le ts'ebetso ho boenjiniere ba software. Sehloohong sena, re tla hlahloba kamoo DSPy e ka nolofatsang ts'ebetso ea tlhahlobo ea likopo tse hahiloeng holim'a LLMs, le ho buisana ka melemo le mathata a paradigm ena e hlahang.

DSPy ke eng?

DSPy ke laebrari ea Python e fanang ka lisebelisoa tse ngata tsa ho iketsetsa tlhahlobo ea software. E lumella bahlahisi ho theha le ho etsa linyeoe tsa liteko, ho beha leihlo liphetho tsa liteko, le ho sekaseka lintlha tsa liteko. DSPy e etselitsoe ho sebetsa ntle le moroallo le LLMs, e nolofalletsang ho theha meralo e tsoetseng pele ea liteko.

Li-LLM ke eng?

Li-LLM ke mofuta oa bohlale ba maiketsetso (AI) bo khonang ho sebetsana le ho hlahisa puo e kang ea batho. Ba koetlisoa ka bongata bo bongata ba lintlha tsa mongolo 'me ba ka ithuta ho lemoha mekhoa, likamano le maemo. Li-LLM li na le lits'ebetso tse ngata ho boenjineri ba software, ho kenyelletsa ts'ebetso ea puo ea tlhaho, tlhahlobo ea mongolo le tlhahiso ea khoutu.

**Ho kopanya DSPy le LLMs **

Ho kopanngoa ha DSPy le LLM ho thusa ho theha linyeoe tsa liteko tse ikemetseng tse ka phethahatsang maemo a rarahaneng, ho etsisa litšebelisano tsa basebelisi, le ho sekaseka liphetho tsa liteko. Khokahano ena e fana ka melemo e mengata, ho kenyelletsa:

Kakaretso e ntlafalitsoeng ea liteko: DSPy e ka hlahisa linyeoe tsa liteko tse akaretsang mefuta e mengata ea maemo, ho netefatsa hore ts'ebeliso e lekoa ka botlalo.
Keketso e sebetsang: Ho itlhahloba ka DSPy le LLM ho fokotsa nako le matsapa a hlokahalang ho phethahatsa linyeoe tsa liteko, ho lokolla bahlahisi hore ba tsepamise maikutlo mesebetsing e meng.
Botšepehi bo ntlafetseng: DSPy e ka beha leihlo liphetho tsa liteko le ho hlokomelisa ba ntlafatsang litaba kapa liphoso life kapa life, ho netefatsa hore sesebelisoa sea tšepahala ebile se tsitsitse.

**Mohlala oa Khoutu: Ho Iketsetsa Linyeoe tsa Teko ka DSPy le LLMs **

Mona ke mohlala oa "code block" o bonts'ang mokhoa oa ho sebelisa DSPy le LLM ho iketsetsa linyeoe tsa liteko:

kenya dspy
ho tsoa ho li-transformer li kenya AutoModelForSeq2SeqLM, AutoTokenizer


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

# Theha nyeoe ea tlhahlobo ea DSPy
teko_case = dspy.TestCase(
    lebitso="Tlhahlobo ea 1",
    tlhaloso="Lekola tshebetso ya ho kena",
    mehato=[
        dspy.Step("Bula leqephe la ho kena"),
        dspy.Step("Kenya lintlha tse nepahetseng"),
        dspy.Step("Tobetsa konopo ea ho kena"),
    ]
)

# Phetha nyeoe ea teko u sebelisa LLM
test_results = dspy.execute_test_case(test_case, model, tokenizer)

# Sekaseka liphetho tsa liteko
test_results.analyse()

Sebaka sena sa khoutu se bontša mokhoa oa ho theha nyeoe ea teko ea DSPy, ho kenya mohlala oa LLM le tokenizer, ho phethahatsa nyeoe ea teko ho sebelisa LLM, le ho sekaseka liphetho tsa liteko.

Kaholimo ho 5-10 Litaelo/Lipotlana tse Sebelisitsoeng ka ho Fetisisa

Mona ke tse 5-10 tse sebelisoang ka ho fetesisa litaelo/likotoana tsa DSPy le LLMs:

dspy.TestCase(): E etsa teko e ncha ea DSPy.
dspy.Step(): E eketsa mohato o mocha tekong ea teko ea DSPy.
dspy.execute_test_case(): E phethahatsa teko ea DSPy e sebelisa LLM.
model = AutoModelForSeq2SeqLM.from_pretrained("t5-base"): E kenya mofuta oa LLM.
tokenizer = AutoTokenizer.from_pretrained("t5-base"): E kenya tokenizer ea LLM.
test_results.analyze(): E sekaseka sephetho sa nyeoe ea teko ea DSPy.
dspy.monitor_test_results(): E beha leihlo liphetho tsa nyeoe ea teko ea DSPy.
dspy.alert_developer(): E lemosa mohlahisi ka litaba kapa liphoso life kapa life.
dspy.generate_test_cases(): E hlahisa linyeoe tse ncha tsa liteko ka LLM.
dspy.optimize_test_cases(): E matlafatsa ts'ebetso ea linyeoe tsa liteko ka LLM.

Pono ya pono

Mona ke pono ea kopanyo ea DSPy le LLM ho sebelisoa Chart.js:

{
  "type": "bar",
  "data": {
    "labels": ["Tlhahlobo ea 1", "Tlhahlobo ea 2", "Tlhahlobo ea 3"],
    "lintlha tsa boitsebiso": [
      {
        "label": "Nako ea Phethahatso ea Nyeoe ea Teko",
        "data": [10, 20, 30],
        "backgroundColor": "rgba(255, 99, 132, 0.2)",
        "borderColor": "rgba(255, 99, 132, 1)",
        "borderWidth": 1
      },
      {
        "label": "Sekhahla sa ho Hloka Teko",
        "data": [5, 10, 15],
        "backgroundColor": "rgba(54, 162, 235, 0.2)",
        "borderColor": "rgba(54, 162, 235, 1)",
        "borderWidth": 1
      }
    ]
  },
  "likgetho": {
    "sehlooho": {
      "bontša": 'nete,
      "text": "Khokahano ea DSPy le LLM"
    },
    "makala": {
      "yAxes": [
        {
          "li-ticks": {
            "beginAtZero": 'nete
          }
        }
      ]
    }
  }
}

Pontšo ena e bonts'a nako ea ts'ebetso le sekhahla sa ho hloleha ha linyeoe tse tharo tsa liteko tse entsoeng ho sebelisoa DSPy le LLM.

Qetello

Ho kopanngoa ha DSPy le LLMs ho fetotse boemo ba tlhahlobo ea software, ho nolofalletsa ho theoa ha linyeoe tsa liteko tse ikemetseng tse ka hlahisang maemo a rarahaneng, tsa etsisa litšebelisano tsa basebelisi, le ho sekaseka liphetho tsa liteko. Ka ho phahamisa bokhoni ba DSPy le LLMs, bahlahisi ba ka nolofatsa ts'ebetso ea tlhahlobo, ba matlafatsa ts'epo le ts'ebetso e ntle, le ho ntlafatsa boleng ba lits'ebetso tsa bona kaofela. Ha theknoloji ena e ntse e tsoela pele ho fetoha, re ka lebella ho bona lits'ebetso tse ncha tsa DSPy le LLM ho boenjineri ba software.

Laola mohlala oa LLM le tokenizer

¿Tienes preguntas?