Mise en forme des sorties (JSON, XML, Markdown)

Pourquoi le format est important

Mode JSON

response = client.chat.completions.create(
    model='gpt-4o-mini',
    messages=messages,
    response_format={'type': 'json_object'},
)
import json
data = json.loads(response.choices[0].message.content)

Sorties structurées (schéma strict)

schema = {
    'name': 'extract_person',
    'schema': {
        'type': 'object',
        'properties': {
            'name': {'type': 'string'},
            'age': {'type': 'integer'}
        },
        'required': ['name', 'age'],
        'additionalProperties': False
    },
    'strict': True
}

response = client.chat.completions.create(
    model='gpt-4o-2024-08-06',
    messages=messages,
    response_format={'type': 'json_schema', 'json_schema': schema}
)

Balises XML (style Anthropic)

system = '''
Return your answer wrapped in XML tags:

<reasoning>Your step-by-step thinking</reasoning>
<answer>The final answer</answer>
'''

# Parse with a simple regex or BeautifulSoup
import re
ans = re.search(r'<answer>(.*?)</answer>', text, re.S).group(1)

Markdown pour les humains

Choisissez le bon format selon le destinataire

Montrez plutôt que de vous contenter d'expliquer

system = '''
Return a JSON object like:
{
  "action": "reply",
  "content": "Hi there!",
  "confidence": 0.95
}
'''
print(system.strip())

Forcez le premier jeton

messages = [
    {'role': 'user', 'content': 'Return JSON with name and age.'},
    {'role': 'assistant', 'content': '{'}
]
# Output will start at '{ "name": ...' guaranteed.
for m in messages:
    print(f"{m['role']}: {m['content']}")
print('Output will start at \'{ "name": ...\' guaranteed.')

Gérez le texte parasite final

Instructions de réparation

repair_prompt = f'''
The previous output failed to parse with error: {error}.
Return ONLY a valid JSON object matching the schema. No prose.

Previous output:
{bad_output}
'''

Pydantic pour la validation

from pydantic import BaseModel

class Person(BaseModel):
    name: str
    age: int

person = Person.model_validate_json(response_text)
print(person.name, person.age)

Meilleur format pour le code

Récapitulatif