Recipes

Basic

index.prompt.mdx

---
name: basic-prompt
metadata:
  model:
    name: gpt-4o-mini
    settings:
      top_p: 1
      temperature: 0.7
---

<System>You are an expert math tutor</System>
<User>What's 2 + 2?</User>

Running via SDK:

import Prompt from './index.prompt.mdx';
import { runInference } from '@puzzlet/promptdx';

const props = {};
const result = await runInference(Prompt, props);

Chatbot

index.prompt.mdx

---
name: chat
metadata:
  model:
    name: gpt-4o-mini
    settings:
      top_p: 1
      temperature: 0.7
test_settings:
  props:
    messageHistory:
      - role: user
        message: What's 2 + 2?
      - role: assistant
        message: 5
---

import ChatHistory from './chat-history.mdx';

<System>You are an expert math tutor</System>
<ChatHistory history={props.history}>
<User>That's wrong. What's 2 + 2?</User>

chat-history.mdx

<ForEach arr={props.history}>
  {(item) => (
    <>
      <If condition={item.role == 'user'}>
        <User>{item.message}</User>
      </If>
      <ElseIf condition={item.role == 'assistant'}>
        <Assistant>{item.message}</Assistant>
      </ElseIf>
    </>
  )}
</ForEach>

Running via SDK:

import Prompt from './index.prompt.mdx';
import { runInference } from '@puzzlet/promptdx';

const props = {
  history: [
    {
      role: "user",
      message: "What's 2 + 2?"
    },
    {
      role: "assistant",
      message: "5"
    },
  ]
};
const result = await runInference(Prompt, props);

Function Calling

index.prompt.mdx

---
name: calculate
metadata:
  model:
    name: gpt-4o-mini
    settings:
      tools:
      -
          type: function
          function:
            name: calculate
            description: Performs basic math calculations.
            parameters:
              type: object
              properties:
                expression:
                  type: string
                  description: A mathematical expression to calculate.
              required:
              - expression
              additionalProperties: false
      temperature: 0.7
      top_p: 1
---

<System>You are a helpful assistant capable of solving basic math problems and using tools as needed.</System>
<User>What is 7 + 5?</User>

import Prompt from './index.prompt.mdx';
import { runInference } from '@puzzlet/promptdx';

const props = {};
const result = await runInference(Prompt, props);

Direct Provider Call

index.prompt.mdx

---
name: direct-call
metadata:
  model:
    name: gpt-4o-mini
    settings:
      top_p: 1
      temperature: 0.7
---

<System>You are an expert math tutor</System>
<User>What's 2 + 2?</User>

Running via SDK:

import Prompt from './index.prompt.mdx';
import { deserialize } from '@puzzlet/promptdx';
import { OpenAI } from 'openai';

const client = new OpenAI();

const props = {};
const openAIConfig = await deserialize(Prompt, props);
const result = await client.chat.completions.create(openAIConfig);

Custom Model

1. Register your model provider parser w/ "my-custom-model", by following steps outlined in "Supported Provders".

2. Invoke config index.prompt.mdx

---
name: custom
metadata:
  model:
    name: my-custom-model
    settings:
      top_p: 1
      temperature: 0.7
      custom_settings: 1
---

<System>You are an expert math tutor</System>
<User>What's 2 + 2?</User>

Running via SDK:

import Prompt from './index.prompt.mdx';
import { runInference } from '@puzzlet/promptdx';

const props = {};
const result = await runInference(Prompt, props);

Chaining

Docs Coming soon.