MIPRO Offline API Reference

Note: This page is auto-generated from SDK validation code. Parameters and types are extracted automatically and will update when the code changes.

---

MIPRO Offline (Multi-prompt Instruction Proposal Optimizer) is an algorithm for optimizing prompts through systematic instruction proposal and evaluation in offline mode, where the backend orchestrates all rollouts by calling your task app.

Endpoint: POST /api/policy-optimization/online/jobs

Authentication: Bearer token via Authorization: Bearer $SYNTH_API_KEY

Overview

In offline mode:

• Backend orchestrates: Backend calls your task app for all rollouts
• Tunneling required: Task app must be publicly accessible
• Automatic evaluation: Backend handles all rollout execution and reward collection
• No local code needed: Just provide task app URL and configuration

Request

{
  "policy_optimization": {
    "algorithm": "mipro",
    "task_app_url": "https://your-task-app.example.com",
    "policy": {
      "model": "gpt-4o-mini",
      "provider": "openai",
      "temperature": 0.0,
      "max_completion_tokens": 256
    },
    "mipro": {
      "mode": "offline",
      "bootstrap_train_seeds": [0, 1, 2, 3, 4],
      "val_seeds": [100, 101, 102],
      "proposer": {
        "mode": "instruction_only",
        "model": "gpt-4o-mini",
        "provider": "openai",
        "temperature": 0.7,
        "max_tokens": 512
      }
    }
  }
}

Parameters

Parameter	Type	Required	Description
mipro.mode	string	Yes	Must be "offline"
mipro.bootstrap_train_seeds	array[int]	Yes	Initial training seeds for bootstrap phase
mipro.val_seeds	array[int]	Yes	Validation seeds for evaluation
mipro.proposer	object	Yes	Proposer configuration for generating prompt proposals
mipro.proposer.max_tokens	int	No	Maximum tokens for proposer output (default: 512)
mipro.proposer.mode	string	Yes	Proposer generation mode: ‘instruction_only’
mipro.proposer.model	string	Yes	Model for generating proposals
mipro.proposer.provider	string	Yes	Provider for proposer model
mipro.proposer.temperature	float	No	Temperature for proposer generation (default: 0.7)
task_app_url	string	Yes	Publicly accessible URL of your task app

Workflow

1. Set up task app: Ensure your task app is publicly accessible (use Cloudflare tunnel for local development)
2. Register API key: Register ENVIRONMENT_API_KEY with the backend
3. Create job: Submit MIPRO job with mode: "offline" and task_app_url
4. Backend orchestrates: Backend automatically:
   • Calls your task app for rollouts
   • Collects rewards
   • Generates new prompt proposals
   • Evaluates candidates
5. Poll for results: Check job status until completion

Response

{
  "job_id": "pl_abc123",
  "status": "running"
}

Polling for Completion

Use GET /api/policy-optimization/online/jobs/{job_id} to check status:

{
  "job_id": "pl_abc123",
  "status": "succeeded",
  "best_score": 0.875
}

Notes

• Tunneling required: Task app must be publicly accessible (use Cloudflare tunnel for local development)
• Environment API key: ENVIRONMENT_API_KEY must be registered with the backend
• Backend orchestrates: Backend handles all rollout execution and reward collection automatically
• No local rollout code: You don’t need to write rollout loops - backend handles everything
• Proposer API key: Automatically resolved from backend environment (OPENAI_API_KEY or PROD_OPENAI_API_KEY)

See Also

• MIPRO Online API - Online mode documentation
• MIPRO SDK Reference - SDK usage guide
• Policy Optimization API - General policy optimization documentation