The Fireworks AI Cookbook provides ready-to-run recipes and utilities for training models on Fireworks. It covers supervised fine-tuning (SFT), reinforcement learning (GRPO, DAPO, GSPO, CISPO), and preference optimization (DPO, ORPO) — all driven by the Fireworks Training SDK.
For full SDK documentation, see the Fireworks Training SDK Reference.
Head to the training/ directory for installation instructions, recipe configuration, and runnable examples.
training/ Training SDK recipes, utilities, and examples
recipes/ Fork-and-customize training loop scripts
utils/ Shared config, data loading, losses, metrics
examples/ Worked examples (e.g. deepmath GRPO)
tests/ Unit and end-to-end tests
archived/ Legacy cookbook content (see below)
All previous cookbook material — learning tutorials, integration examples, showcase projects, evaluation recipes, and more — has been moved to archived/. See the archived README for details on what's there.
We welcome contributions! See the Contribution Guide for how to get started.