Strukturéiert Output

POST https://api.shannon-ai.com/v1/chat/completions

import json
from openai import OpenAI

client = OpenAI(api_key="YOUR_API_KEY", base_url="https://api.shannon-ai.com/v1")

response = client.chat.completions.create(
    model="Kimi-K3-3BIT-REAP",
    messages=[
        {"role": "user", "content": "Extract the person: John Doe, 30 years old, engineer."}
    ],
    response_format={
        "type": "json_schema",
        "json_schema": {
            "name": "person",
            "strict": True,
            "schema": {
                "type": "object",
                "properties": {
                    "name": {"type": "string"},
                    "age": {"type": "integer"},
                    "occupation": {"type": "string"},
                },
                "required": ["name", "age", "occupation"],
            },
        },
    },
)

# The JSON arrives as text in message.content: parse it, then check it.
try:
    person = json.loads(response.choices[0].message.content)
except json.JSONDecodeError:
    person = None  # ask again, or report the failure

print(person)

import OpenAI from "openai";

const client = new OpenAI({ apiKey: "YOUR_API_KEY", baseURL: "https://api.shannon-ai.com/v1" });

const response = await client.chat.completions.create({
  model: "Kimi-K3-3BIT-REAP",
  messages: [
    { role: "user", content: "Extract the person: John Doe, 30 years old, engineer." },
  ],
  response_format: {
    type: "json_schema",
    json_schema: {
      name: "person",
      strict: true,
      schema: {
        type: "object",
        properties: {
          name: { type: "string" },
          age: { type: "integer" },
          occupation: { type: "string" },
        },
        required: ["name", "age", "occupation"],
      },
    },
  },
});

// The JSON arrives as text in message.content: parse it, then check it.
let person = null;
try {
  person = JSON.parse(response.choices[0].message.content);
} catch {
  // ask again, or report the failure
}

console.log(person);

curl https://api.shannon-ai.com/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "Kimi-K3-3BIT-REAP",
    "messages": [
      {"role": "user", "content": "Extract the person: John Doe, 30 years old, engineer."}
    ],
    "response_format": {
      "type": "json_schema",
      "json_schema": {
        "name": "person",
        "strict": true,
        "schema": {
          "type": "object",
          "properties": {
            "name": {"type": "string"},
            "age": {"type": "integer"},
            "occupation": {"type": "string"}
          },
          "required": ["name", "age", "occupation"]
        }
      }
    }
  }'

D'JSON ass den Text vun der Äntwert, an choices[0].message.content:

{
  "id": "chatcmpl-0c4e7a1b9d2f4e6a8b3c5d7e9f1a2b3c",
  "object": "chat.completion",
  "created": 1791590400,
  "model": "Kimi-K3-3BIT-REAP",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "{\"name\":\"John Doe\",\"age\":30,\"occupation\":\"engineer\"}",
        "reasoning_content": null
      },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 96,
    "completion_tokens": 19,
    "total_tokens": 115,
    "prompt_tokens_details": {"cached_tokens": 0},
    "completion_tokens_details": {"reasoning_tokens": 0}
  }
}

D'Moduse vun response_format

response_format ass e Feld vun /v1/chat/completions. Ouni et ass d'Äntwert normalen Text.

De JSON-Objet-Modus brauch kee Schema. Nennt d'Schlësselen am Prompt:

{
  "model": "shannon-3",
  "messages": [
    {
      "role": "user",
      "content": "Name the capital of France. Answer as JSON with the keys city and country."
    }
  ],
  "response_format": {"type": "json_object"}
}

Ënnerstëtzung pro Modell

GET /v1/models mellt datselwecht pro ID: capabilities.json_schema bei all Modell, an capabilities.response_format bei den Hosted Open-Weight-Modeller. Modeller a Präisser

Ënnerstëtzt Schema-Schlësselwierder

Wat e Modell vun Ärem Schema kritt, hänkt vum Modell an vum Endpoint of:

D'Ënnermengt u Schlësselwierder vun de Shannon-Modeller op /v1/chat/completions:

Op /v1/responses: text.format

/v1/responses schreift déiselwecht Ufro als text.format. D'Schema steet direkt an format, niewent name an strict, net ënner engem json_schema-Schlëssel. Fir de JSON-Objet-Modus schéckt {"format": {"type": "json_object"}}.

{
  "model": "Kimi-K3-3BIT-REAP",
  "input": "Extract the person: John Doe, 30 years old, engineer.",
  "text": {
    "format": {
      "type": "json_schema",
      "name": "person",
      "strict": true,
      "schema": {
        "type": "object",
        "properties": {
          "name": {"type": "string"},
          "age": {"type": "integer"},
          "occupation": {"type": "string"}
        },
        "required": ["name", "age", "occupation"]
      }
    }
  }
}

D'JSON ass den text vum output_text-Deel vum message-Item:

{
  "output": [
    {
      "id": "msg_7e1d3c5b9a2f4d6e8c0b1a3f5e7d9c2b",
      "type": "message",
      "role": "assistant",
      "status": "completed",
      "content": [
        {
          "type": "output_text",
          "text": "{\"name\":\"John Doe\",\"age\":30,\"occupation\":\"engineer\"}",
          "annotations": []
        }
      ]
    }
  ]
}

D'Ënnerstëtzung pro Modell an d'Feeler sinn déiselwecht wéi op /v1/chat/completions. Responses-API

Op /v1/messages: e Tool als Schema

D'Messages-Format huet kee Feld fir strukturéierten Output. Fir JSON vun deene selwechte Modeller rufft /v1/chat/completions oder /v1/responses mam selwechte Schlëssel.

Fir op /v1/messages ze bleiwen, beschreift d'Objet als Tool: setzt Äert Schema an input_schema an nennt d'Tool an tool_choice. Den input vum tool_use-Block ass dann Äert Objet, schonn geparst. E genannten tool_choice gëtt vun den Hosted Open-Weight-Modeller ugewannt.

{
  "model": "Kimi-K3-3BIT-REAP",
  "max_tokens": 1024,
  "messages": [
    {
      "role": "user",
      "content": "Extract the person: John Doe, 30 years old, engineer."
    }
  ],
  "tools": [
    {
      "name": "record_person",
      "description": "Record the extracted person.",
      "input_schema": {
        "type": "object",
        "properties": {
          "name": {"type": "string"},
          "age": {"type": "integer"},
          "occupation": {"type": "string"}
        },
        "required": ["name", "age", "occupation"]
      }
    }
  ],
  "tool_choice": {"type": "tool", "name": "record_person"}
}

{
  "stop_reason": "tool_use",
  "content": [
    {
      "type": "tool_use",
      "id": "toolu_4b8d2f6a1c3e4a5b9d7f0e2c4a6b8d1f",
      "name": "record_person",
      "input": {"name": "John Doe", "age": 30, "occupation": "engineer"}
    }
  ]
}

Kontrolléiert, datt stop_reason tool_use ass, ier Dir de Block liest. Funktiounsofruff

Wéi d'JSON ukënnt

• D'JSON ass e String. D'Äntwert huet kee geparst Objet derniewent: loosst Äre JSON-Parser op message.content lafen.
• Mat stream: true kënnt d'JSON a Stécker vun delta.content. Setzt se zesummen a parst eemol, wann de Stream eriwwer ass; e Stéck eleng ass kee gültegt JSON.
• En Hosted Open-Weight-Modell beäntwert eng response_format-Ufro ouni Reasoning-Text: reasoning_content ass null.

Kontrolléiert, wat Dir kritt

D'Schema leet fest, wéi de Modell schreift. D'API vergläicht déi fäerdeg Äntwert net mat Ärem Schema, Äre Code mécht dofir déi lescht Kontroll:

• Parst an engem Error-Handler, wéi d'Beispill uewen.
• Validéiert dat geparst Objet mam selwechte Schema an Ärem eegene Code, zum Beispill mat enger JSON-Schema-Bibliothéik, Pydantic oder Zod.
• Eng laang Äntwert ka ophalen, ier d'JSON komplett ass. Eng Äntwert, déi net gestreamt gëtt, mellt finish_reason nëmmen als stop oder tool_calls, dofir ass e Parse-Feeler d'Signal. Freet nach eng Kéier, oder freet e méi klengt Objet.

Feeler

Déi zwou 400-Äntwerte ginn geschéckt, ier eppes aus Ärem Solde ofgezunn gëtt. Feelerbehandlung