Fireworks AI Cookbook

Ready-to-run training recipes for reinforcement learning (GRPO, DAPO, GSPO, CISPO), preference optimization (DPO, ORPO), and supervised fine-tuning (SFT) on Fireworks.

Full documentation: Fireworks Training API

Quick Start

git clone https://github.com/fw-ai/cookbook.git
cd cookbook/training
conda create -n cookbook python=3.12 -y && conda activate cookbook
pip install --pre -e .

See training/README.md for configuration, recipes, and examples.

For AI Agents

One canonical skill covers the full training product: skills/fireworks-training/SKILL.md. It routes managed SFT/DPO/ORPO/RFT and Training API serverless or dedicated workflows, then progressively loads the relevant cookbook, operations, or troubleshooting references.

Claude Code

claude plugin marketplace add fw-ai/cookbook
claude plugin install fireworks-training@fw-ai-cookbook

Cursor

npx --yes skills add fw-ai/cookbook -g -s fireworks-training -a cursor -y

Codex

npx --yes skills add fw-ai/cookbook -g -s fireworks-training -a codex -y

The repository also includes .codex-plugin/plugin.json for packaging the same skill as a Codex plugin. The skill is portable Agent Skills Markdown; Cursor and Codex installation is validated with the skills CLI and is not limited to the Claude compact interface. firectl may still require mutating commands to be run manually in the user's terminal when its AI-agent safety guard is active.

Repository Structure

training/ is the primary development surface. eval/ contains reproducible evaluation packages, and tools/ holds standalone customer-facing scripts (public firectl workflows). Legacy integrations, multimedia examples, and earlier cookbook content live under archived/.

training/           Training API recipes, utilities, and examples
  recipes/          Fork-and-customize training loop scripts
  utils/            Shared config, data loading, losses, metrics
  examples/         Worked examples (RL, SFT, DPO, ORPO)
  renderer/         Local renderers and correctness verifier
  tests/            Unit and end-to-end tests
eval/               Reproducible evaluation packages and benchmark adapters
tools/              Standalone customer scripts (downloadable from GitHub)
skills/             One Fireworks training skill and progressive references
archived/           Legacy integrations, multimedia, and cookbook content

Evaluations

eval/healthbench_professional/ — run OpenAI's HealthBench Professional through Harbor, preserve exact Fireworks input/output token IDs and behavior-policy logprobs, and export validated trajectories for RL workflows.

Training skill

skills/fireworks-training/SKILL.md — install once, then ask your coding agent to plan, run, resume, deploy, or debug training. The skill uses public firectl for managed jobs and the cookbook for custom Training API loops.

Contributing

See the Contribution Guide.

fireworks-training

Popularity

What's Inside

README