Debug and migrate training work on Fireworks via the cookbook. Covers greenfield work (pick a recipe, fork it, resolve training + deployment shape from a profile), distillation / OPD / SDFT recipe work, user-level recovery (promote a checkpoint, list promotable checkpoints on a trainer, self-check a `WeightSyncScope.PER_TRAINER` vs `PER_DEPLOYMENT` bucket-scope mix-up), and migrating deprecated managed-infra scripts (port `InfraConfig` / `setup_infra` / `ResourceCleanup` to `TrainerConfig` + the recipe's SDK-managed provisioning path). The cookbook is the reference implementation of `fireworks.training.sdk`; fork a recipe or run an example instead of reimplementing. Trigger when the user wants to start, resume, promote, distill, migrate off deprecated infra APIs, or do a first-line diagnosis on a training run; for deeper recovery the skill routes users to Fireworks support.
Use public Fireworks firectl for account inspection and customer-controlled LoRA SFT tuning workflows.
Run end-to-end fine-tuning on Fireworks via the Fireworks Agent (a.k.a. Pilot Agent, exposed through `firectl session`). Use this skill whenever the user asks to fine-tune, SFT, DPO, ORPO, post-train, custom-train, auto-train, hyperparameter-tune, or hp-sweep a model on Fireworks — including `qwen3-4b`, `qwen3-8b`, `qwen3p5-9b`, `qwen3p5-27b`, `qwen3p5-35b-a3b`, `llama-v3p3-70b-instruct`, `kimi-k2p5`, `kimi-k2p6`, `glm-5p1`, `qwen3p5-397b-a17b`, or any other Fireworks-hosted base model. Also use it for kicking off a training session, monitoring or resuming a session, answering the agent's mid-run questions, approving training plans, troubleshooting `failed` sessions, listing/cancelling/deleting sessions, and figuring out how to call a fine-tuned model after training. Triggers on commands like `firectl session create / get / events / list / update / cancel / delete`, on terms like "Pilot Agent", "Fireworks Agent", "auto-tune on Fireworks", "fine-tune on Fireworks", and on session IDs of the form `1779157376-41