Training loops and worked examples for RL (GRPO, IGPO, SAO/PPO), preference optimization (DPO, ORPO), SFT, distillation and embedding fine-tuning on Fireworks. Full docs: Training API.
git clone https://github.com/fw-ai/cookbook.git
cd cookbook/training
uv venv --python 3.12 && source .venv/bin/activate
uv pip install -e .See training/README.md for recipes and examples.
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 plugin marketplace add fw-ai/cookbook
claude plugin install fireworks-training@fw-ai-cookbooknpx --yes skills add fw-ai/cookbook -g \
-s fireworks-training -s research -s configure -s debug -a cursor -ynpx --yes skills add fw-ai/cookbook -g \
-s fireworks-training -s research -s configure -s debug -a codex -yThe 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.
training/ Training API cookbook (primary surface)
recipes/ One training loop per file; fork to change the algorithm
utils/ Pieces recipes assemble: config, data, losses, rollout, metrics
examples/ Task code + one recipe call (dedicated and serverless twins)
case-studies/ End-to-end notebooks
eval/ Reproducible evaluation packages and benchmark adapters
skills/ Research, configure and debug agent workflows
archived/ Legacy integrations and earlier cookbook content
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.
See the Contribution Guide.