Structured outputs

When using the .create method in the Python SDK, we can generate structured outputs if we provide a pydantic model with our prompt.

When using .create method in Typescript SDK, we can generate structured outputs if we provide a Zod object with our prompt.

from notdiamond import NotDiamond
from pydantic import BaseModel, Field

class LanguageChoice(BaseModel):
    language: str = Field(description="The programming language of choice.")
    reason: str = Field(description="The reason to pick the programming language.")

messages = [
        {"role": "system", "content": "You are a world class software developer."},
        {"role": "user", "content": "What language would you suggest for developing a web application?"}
    ]

llm_providers = ['openai/gpt-3.5-turbo', 'openai/gpt-4-turbo-2024-04-09', 
                 'openai/gpt-4o-2024-05-13', 'anthropic/claude-3-opus-20240229']

client = NotDiamond()

result, session_id, _ = client.chat.completions.create(
        messages=messages,
        model=llm_providers,
        response_model=LanguageChoice
    )

print(result)


import { z } from "zod";

const structuredOutput = z.object({
  name: z.string(),
  age: z.union([z.number(), z.string()]),
});

const notDiamond = new NotDiamond();

const result = await notDiamond.create({
  messages: [{ content: 'What is 12x12?', role: 'user' }],
  llmProviders: [
    { provider: 'openai', model: 'gpt-4o-2024-05-13' },
    { provider: 'anthropic', model: 'claude-3-5-sonnet-20240620' },
    { provider: 'google', model: 'gemini-1.5-pro-latest' },
  ],
  tradeoff: 'latency',
  responseModel: structuredOutput,
});

We can also do this with a streamed response. Response model fields will start with None and gradually build out as the response is streamed. Example:

from notdiamond import NotDiamond
from pydantic import BaseModel, Field

class LanguageChoice(BaseModel):
    language: str = Field(description="The programming language of choice.")
    reason: str = Field(description="The reason to pick the programming language.")

messages = [
        {"role": "system", "content": "You are a world class software developer."},
        {"role": "user", "content": "What language would you suggest for developing a web application?"}
    ]

llm_providers = ['openai/gpt-3.5-turbo', 'openai/gpt-4-turbo-2024-04-09', 
                 'openai/gpt-4o-2024-05-13', 'anthropic/claude-3-opus-20240229']

client = NotDiamond(llm_configs=llm_providers)

result = client.chat.completions.stream(
        messages=messages,
        response_model=LanguageChoice
    )
for chunk in result:
    print(chunk)

import { z } from "zod";

const structuredOutput = z.object({
  name: z.string(),
  age: z.union([z.number(), z.string()]),
});

const notDiamond = new NotDiamond();

const result = await notDiamond.stream({
  messages: [{ content: 'What is 12x12?', role: 'user' }],
  llmProviders: [
    { provider: 'openai', model: 'gpt-4o-2024-05-13' },
    { provider: 'anthropic', model: 'claude-3-5-sonnet-20240620' },
    { provider: 'google', model: 'gemini-1.5-pro-latest' },
  ],
  tradeoff: 'latency',
  responseModel: structuredOutput,
});

for await (const chunk of result.stream) {
  setMessage((prevMessage) => prevMessage + chunk);
}

📘
Not all models support structured outputs
Visit the supported LLM models page to see which models support structured outputs.

📘Not all models support structured outputs

📘
Not all models support structured outputs