Skip to main content
Write custom Python recipes for training, evaluation, and other AI workflows using adaptive_harmony.

​
Getting Started

Installation

Install adaptive-harmony

Local Development

Connect to compute plane for interactive development

Recipe Syntax

Structure and entry points

Configuration

Define recipe parameters

​
Building Recipes

Models

Load and spawn models

Datasets

Load datasets and StringThread

Graders

Custom reward functions

Platform Graders

Load graders from Adaptive

​
Training & Evaluation

Training Recipes

SFT, PPO, GRPO, DPO algorithms

Evaluation Recipes

Evaluate models and create artifacts

Structured Output

JSON and constrained generation

Progress Reporting

Track recipe progress in UI

​
Quick Reference

Upload a recipe: 
adaptive.recipes.upload(
    path="/path/to/recipe.py",  # or directory
    recipe_key="my-recipe",
    name="My Training Recipe"
)
When uploading a directory, the entry point must be main.py and all imports must be relative (e.g., from .utils import helper).
Run a recipe: 
job = adaptive.jobs.run(
    recipe_key="my-recipe",
    num_gpus=2,
    args={"model": "llama-3.1-8b", "dataset": "my-data"}
)