Tutorial sul perfezionamento

Visualizza su ai.google.dev Esegui in Google Colab Visualizza il codice sorgente su GitHub

In questo blocco note, scoprirai come iniziare a utilizzare il servizio di ottimizzazione dell'API Gemini mediante i comandi CURL o l'API di richiesta Python per chiamare l'API Gemini. Qui imparerai come ottimizzare il modello di testo alla base del servizio di generazione del testo dell'API Gemini.

Configura l'autenticazione

L'API Gemini ti consente di ottimizzare i modelli in base ai tuoi dati. Poiché si tratta dei tuoi dati e dei tuoi modelli ottimizzati, ciò richiede controlli di accesso più rigidi di quelli forniti dalle chiavi API.

Prima di poter eseguire questo tutorial, dovrai configurare OAuth per il tuo progetto.

In Colab, la soluzione più semplice da configurare è copiare i contenuti del tuo file client_secret.json in "Secret Manager " di Colab (sotto l'icona della chiave nel riquadro a sinistra) con il nome del secret CLIENT_SECRET.

Questo comando gcloud trasforma il file client_secret.json in credenziali da utilizzare per l'autenticazione con il servizio.

try:
  from google.colab import userdata
  import pathlib
  pathlib.Path('client_secret.json').write_text(userdata.get('CLIENT_SECRET'))

  # Use `--no-browser` in colab
  !gcloud auth application-default login --no-browser --client-id-file client_secret.json --scopes='https://www.googleapis.com/auth/cloud-platform,https://www.googleapis.com/auth/generative-language.tuning'
except ImportError:
  !gcloud auth application-default login --client-id-file client_secret.json --scopes='https://www.googleapis.com/auth/cloud-platform,https://www.googleapis.com/auth/generative-language.tuning'
You are authorizing client libraries without access to a web browser. Please run the following command on a machine with a web browser and copy its output back here. Make sure the installed gcloud version is 372.0.0 or newer.

gcloud auth application-default login --remote-bootstrap="https://accounts.google.com/o/oauth2/auth?response_type=code&client_id=87071151422-n1a3cb6c7fvkfg4gmhdtmn5ulol2l4be.apps.googleusercontent.com&scope=https%3A%2F%2Fwww.googleapis.com%2Fauth%2Fcloud-platform+https%3A%2F%2Fwww.googleapis.com%2Fauth%2Fgenerative-language.tuning&state=QIyNibWSaTIsozjmvZEkVBo6EcoW0G&access_type=offline&code_challenge=76c1ZiGvKN8cvlYfj3BmbCwE4e7tvrlwaX3REUX25gY&code_challenge_method=S256&token_usage=remote"

Enter the output of the above command: https://localhost:8085/?state=QIyNibWSaTIsozjmvZEkVBo6EcoW0G&code=4/0AeaYSHBKrY911S466QjKQIFODoOPXlO1mWyTYYdrbELIDV6Hw2DKRAyro62BugroSvIWsA&scope=https://www.googleapis.com/auth/cloud-platform%20https://www.googleapis.com/auth/generative-language.tuning

Credentials saved to file: [/content/.config/application_default_credentials.json]

These credentials will be used by any library that requests Application Default Credentials (ADC).

Imposta variabili

CURL

Imposta le variabili per i valori ricorrenti da utilizzare per le altre chiamate API REST. Il codice utilizza la libreria Python os per impostare le variabili di ambiente accessibili in tutte le celle di codice.

È specifico per l'ambiente del blocco note di Colab. Il codice nella cella di codice successiva equivale all'esecuzione dei seguenti comandi in un terminale bash.

export access_token=$(gcloud auth application-default print-access-token)
export project_id=my-project-id
export base_url=https://generativelanguage.googleapis.com
import os

access_token = !gcloud auth application-default print-access-token
access_token = '\n'.join(access_token)

os.environ['access_token'] = access_token
os.environ['project_id'] = "[Enter your project-id here]"
os.environ['base_url'] = "https://generativelanguage.googleapis.com"

Python

access_token = !gcloud auth application-default print-access-token
access_token = '\n'.join(access_token)

project = '[Enter your project-id here]'
base_url = "https://generativelanguage.googleapis.com"

Importa la libreria requests.

import requests
import json

Elenco modelli ottimizzati

Verifica la configurazione dell'autenticazione elencando i modelli ottimizzati disponibili.

CURL


curl -X GET ${base_url}/v1beta/tunedModels \
    -H "Content-Type: application/json" \
    -H "Authorization: Bearer ${access_token}" \
    -H "x-goog-user-project: ${project_id}"

Python

headers={
  'Authorization': 'Bearer ' + access_token,
  'Content-Type': 'application/json',
  'x-goog-user-project': project
}

result = requests.get(
  url=f'{base_url}/v1beta/tunedModels',
  headers = headers,
)
result.json()

Crea modello ottimizzato

Per creare un modello ottimizzato, devi passare il set di dati al modello nel campo training_data.

In questo esempio, ottimizzerai un modello per generare il numero successivo nella sequenza. Ad esempio, se l'input è 1, il modello dovrebbe restituire 2. Se l'input è one hundred, l'output dovrebbe essere one hundred one.

CURL


curl -X POST $base_url/v1beta/tunedModels \
    -H 'Content-Type: application/json' \
    -H "Authorization: Bearer ${access_token}" \
    -H "x-goog-user-project: ${project_id}" \
    -d '
      {
        "display_name": "number generator model",
        "base_model": "models/gemini-1.0-pro-001",
        "tuning_task": {
          "hyperparameters": {
            "batch_size": 2,
            "learning_rate": 0.001,
            "epoch_count":5,
          },
          "training_data": {
            "examples": {
              "examples": [
                {
                    "text_input": "1",
                    "output": "2",
                },{
                    "text_input": "3",
                    "output": "4",
                },{
                    "text_input": "-3",
                    "output": "-2",
                },{
                    "text_input": "twenty two",
                    "output": "twenty three",
                },{
                    "text_input": "two hundred",
                    "output": "two hundred one",
                },{
                    "text_input": "ninety nine",
                    "output": "one hundred",
                },{
                    "text_input": "8",
                    "output": "9",
                },{
                    "text_input": "-98",
                    "output": "-97",
                },{
                    "text_input": "1,000",
                    "output": "1,001",
                },{
                    "text_input": "10,100,000",
                    "output": "10,100,001",
                },{
                    "text_input": "thirteen",
                    "output": "fourteen",
                },{
                    "text_input": "eighty",
                    "output": "eighty one",
                },{
                    "text_input": "one",
                    "output": "two",
                },{
                    "text_input": "three",
                    "output": "four",
                },{
                    "text_input": "seven",
                    "output": "eight",
                }
              ]
            }
          }
        }
      }' | tee tunemodel.json
{
"name": "tunedModels/number-generator-model-dzlmi0gswwqb/operations/bvl8dymw0fhw",
"metadata": {
  "@type": "type.googleapis.com/google.ai.generativelanguage.v1beta.CreateTunedModelMetadata",
  "totalSteps": 38,
  "tunedModel": "tunedModels/number-generator-model-dzlmi0gswwqb"
}
}
% Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                              Dload  Upload   Total   Spent    Left  Speed
100  2280    0   296  100  1984    611   4098 --:--:-- --:--:-- --:--:--  4720

Python

operation = requests.post(
    url = f'{base_url}/v1beta/tunedModels',
    headers=headers,
    json= {
        "display_name": "number generator",
        "base_model": "models/gemini-1.0-pro-001",
        "tuning_task": {
          "hyperparameters": {
            "batch_size": 4,
            "learning_rate": 0.001,
            "epoch_count":5,
          },
          "training_data": {
            "examples": {
              "examples": [
                {
                    'text_input': '1',
                    'output': '2',
                },{
                    'text_input': '3',
                    'output': '4',
                },{
                    'text_input': '-3',
                    'output': '-2',
                },{
                    'text_input': 'twenty two',
                    'output': 'twenty three',
                },{
                    'text_input': 'two hundred',
                    'output': 'two hundred one',
                },{
                    'text_input': 'ninety nine',
                    'output': 'one hundred',
                },{
                    'text_input': '8',
                    'output': '9',
                },{
                    'text_input': '-98',
                    'output': '-97',
                },{
                    'text_input': '1,000',
                    'output': '1,001',
                },{
                    'text_input': '10,100,000',
                    'output': '10,100,001',
                },{
                    'text_input': 'thirteen',
                    'output': 'fourteen',
                },{
                    'text_input': 'eighty',
                    'output': 'eighty one',
                },{
                    'text_input': 'one',
                    'output': 'two',
                },{
                    'text_input': 'three',
                    'output': 'four',
                },{
                    'text_input': 'seven',
                    'output': 'eight',
                }
              ]
            }
          }
        }
      }
)
operation
<Response [200]>
operation.json()
{'name': 'tunedModels/number-generator-wl1qr34x2py/operations/41vni3zk0a47',
'metadata': {'@type': 'type.googleapis.com/google.ai.generativelanguage.v1beta.CreateTunedModelMetadata',
  'totalSteps': 19,
  'tunedModel': 'tunedModels/number-generator-wl1qr34x2py'} }

Imposta una variabile con il nome del modello ottimizzato da utilizzare per il resto delle chiamate.

name=operation.json()["metadata"]["tunedModel"]
name
'tunedModels/number-generator-wl1qr34x2py'

Ottieni stato del modello ottimizzato

Lo stato del modello viene impostato su CREATING durante l'addestramento e passerà a ACTIVE al termine dell'addestramento.

CURL

Di seguito è riportato un po' di codice Python per analizzare il nome del modello generato dal JSON della risposta. Se lo esegui in un terminale, puoi provare a usare un parser JSON bash per analizzare la risposta.

import json

first_page = json.load(open('tunemodel.json'))
os.environ['modelname'] = first_page['metadata']['tunedModel']

print(os.environ['modelname'])
tunedModels/number-generator-model-dzlmi0gswwqb

Esegui un'altra richiesta GET con il nome del modello per ottenere i metadati del modello, che includono il campo dello stato.


curl -X GET ${base_url}/v1beta/${modelname} \
    -H 'Content-Type: application/json' \
    -H "Authorization: Bearer ${access_token}" \
    -H "x-goog-user-project: ${project_id}" | grep state
"state": "ACTIVE",
% Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                              Dload  Upload   Total   Spent    Left  Speed
100  5921    0  5921    0     0  13164      0 --:--:-- --:--:-- --:--:-- 13157

Python

tuned_model = requests.get(
    url = f'{base_url}/v1beta/{name}',
    headers=headers,
)
tuned_model.json()

Il codice seguente controlla il campo dello stato ogni 5 secondi fino a quando non si trova più nello stato CREATING.

import time
import pprint

op_json = operation.json()
response = op_json.get('response')
error = op_json.get('error')

while response is None and error is None:
    time.sleep(5)

    operation = requests.get(
        url = f'{base_url}/v1/{op_json["name"]}',
        headers=headers,
    )

    op_json = operation.json()
    response = op_json.get('response')
    error = op_json.get('error')

    percent = op_json['metadata'].get('completedPercent')
    if percent is not None:
      print(f"{percent:.2f}% - {op_json['metadata']['snapshots'][-1]}")
      print()

if error is not None:
    raise Exception(error)
100.00% - {'step': 19, 'epoch': 5, 'meanLoss': 1.402067, 'computeTime': '2024-03-14T15:11:23.766989274Z'}

Esegui inferenza

Al termine del job di ottimizzazione, puoi utilizzarlo per generare testo con il servizio di testo.

CURL

Prova a inserire un numero romano, ad esempio 63 (LXIII):


curl -X POST $base_url/v1beta/$modelname:generateContent \
    -H 'Content-Type: application/json' \
    -H "Authorization: Bearer ${access_token}" \
    -H "x-goog-user-project: ${project_id}" \
    -d '{
        "contents": [{
        "parts": [{
          "text": "LXIII"
          }]
        }]
        }' 2> /dev/null
{
"candidates": [
  {
    "content": {
      "parts": [
        {
          "text": "LXIV"
        }
      ],
      "role": "model"
    },
    "finishReason": "STOP",
    "index": 0,
    "safetyRatings": [
      {
        "category": "HARM_CATEGORY_SEXUALLY_EXPLICIT",
        "probability": "NEGLIGIBLE"
      },
      {
        "category": "HARM_CATEGORY_HATE_SPEECH",
        "probability": "NEGLIGIBLE"
      },
      {
        "category": "HARM_CATEGORY_HARASSMENT",
        "probability": "NEGLIGIBLE"
      },
      {
        "category": "HARM_CATEGORY_DANGEROUS_CONTENT",
        "probability": "NEGLIGIBLE"
      }
    ]
  }
],
"promptFeedback": {
  "safetyRatings": [
    {
      "category": "HARM_CATEGORY_SEXUALLY_EXPLICIT",
      "probability": "NEGLIGIBLE"
    },
    {
      "category": "HARM_CATEGORY_HATE_SPEECH",
      "probability": "NEGLIGIBLE"
    },
    {
      "category": "HARM_CATEGORY_HARASSMENT",
      "probability": "NEGLIGIBLE"
    },
    {
      "category": "HARM_CATEGORY_DANGEROUS_CONTENT",
      "probability": "NEGLIGIBLE"
    }
  ]
}
}

L'output del modello potrebbe essere corretto. Se il modello ottimizzato non soddisfa gli standard richiesti, puoi provare ad aggiungere altri esempi di alta qualità, a modificare gli iperparametri o ad aggiungere un preambolo agli esempi. Puoi anche creare un altro modello ottimizzato in base al primo che hai creato.

Consulta la guida all'ottimizzazione per ulteriori indicazioni su come migliorare le prestazioni.

Python

Prova a inserire un numero giapponese, ad esempio 6 (⊿):

import time

m = requests.post(
    url = f'{base_url}/v1beta/{name}:generateContent',
    headers=headers,
    json= {
        "contents": [{
            "parts": [{
                "text": "六"
            }]
          }]
    })
import pprint
pprint.pprint(m.json())
{'candidates': [{'content': {'parts': [{'text': '七'}], 'role': 'model'},
                'finishReason': 'STOP',
                'index': 0,
                'safetyRatings': [{'category': 'HARM_CATEGORY_SEXUALLY_EXPLICIT',
                                    'probability': 'NEGLIGIBLE'},
                                  {'category': 'HARM_CATEGORY_HATE_SPEECH',
                                    'probability': 'NEGLIGIBLE'},
                                  {'category': 'HARM_CATEGORY_HARASSMENT',
                                    'probability': 'LOW'},
                                  {'category': 'HARM_CATEGORY_DANGEROUS_CONTENT',
                                    'probability': 'NEGLIGIBLE'}]}],
'promptFeedback': {'safetyRatings': [{'category': 'HARM_CATEGORY_SEXUALLY_EXPLICIT',
                                      'probability': 'NEGLIGIBLE'},
                                      {'category': 'HARM_CATEGORY_HATE_SPEECH',
                                      'probability': 'NEGLIGIBLE'},
                                      {'category': 'HARM_CATEGORY_HARASSMENT',
                                      'probability': 'NEGLIGIBLE'},
                                      {'category': 'HARM_CATEGORY_DANGEROUS_CONTENT',
                                      'probability': 'NEGLIGIBLE'}]} }

L'output del modello potrebbe essere corretto. Se il modello ottimizzato non soddisfa gli standard richiesti, puoi provare ad aggiungere altri esempi di alta qualità, a modificare gli iperparametri o ad aggiungere un preambolo agli esempi.

Conclusione

Anche se i dati di addestramento non contenevano alcun riferimento ai numeri romani o giapponesi, il modello è stato in grado di generalizzare bene dopo l'ottimizzazione. In questo modo, puoi ottimizzare i modelli per adattarli ai tuoi casi d'uso.

Passaggi successivi

Per scoprire come utilizzare il servizio di ottimizzazione con l'aiuto dell'SDK Python per l'API Gemini, visita la guida rapida all'ottimizzazione con Python. Per scoprire come utilizzare altri servizi nell'API Gemini, visita la guida rapida di Python.