Skip to content

Repository files navigation

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

These skills bring Fireworks training know-how into compatible AI agents through progressive disclosure. Each entry point loads only the workflow guidance needed for the task, then follows linked Fireworks documentation and runnable cookbook examples when deeper detail is needed. The skill set provides three task-specific skills: research, configure, and debug.

Skill What it does Try
Research Helps decide whether training is the right intervention. It gathers the task, data, evaluation criteria, and constraints, then recommends a method and the closest runnable cookbook entry. It does not launch training. "Which cookbook entry fits prompt routing to small vs big models?"
Configure Turns a training goal into an executable plan for managed SFT, DPO, ORPO, or RFT, as well as serverless or dedicated Training API setups. It validates inputs, estimates cost, and supports running, monitoring, evaluation, deployment, resume, and teardown. It shows the complete plan and asks for approval before spend or mutation. "SFT qwen3-8b on my JSONL. Show the plan, but do not start yet."
Debug Diagnoses stuck, failed, slow, or low-quality training runs. It gathers runtime evidence, classifies the failure, suggests the safest next action, and does not retry or mutate resources without approval. "My job is stuck RUNNING at 0%."

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 -s research -s configure -s debug -a cursor -y

Codex

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

The repository also includes .codex-plugin/plugin.json for packaging the skill set as a Codex plugin. The skills use portable Agent Skills Markdown and can be consumed by other compatible agents. The commands above cover the three validated installation paths. 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. Legacy integrations, standalone customer scripts, 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
skills/             Research, configure, and debug agent workflows
archived/           Legacy integrations, multimedia, and cookbook content
  tools/            Archived standalone customer scripts

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.

Contributing

See the Contribution Guide.

Support

Releases

Packages

Used by

Contributors

Languages