Skip to main content

​
What this is

Use managed resources when you want platform-orchestrated training lifecycles with minimal custom loop code.

​
Why this approach

  • Managed jobs are best when objective families are already supported and your priority is reliability, speed, and operational simplicity.
  • Managed objective coverage includes supervised fine-tuning (SFT), DPO, and managed reinforcement fine-tuning jobs.

​
How to use these APIs

  • Fireworks.supervised_fine_tuning_jobs.*: Lifecycle for SFT jobs.
  • Fireworks.reinforcement_fine_tuning_jobs.*: Lifecycle for managed RL fine-tuning jobs.
  • Fireworks.dpo_jobs.*: Lifecycle for DPO jobs.

​
End-to-end examples

​
Inspect managed jobs

sft_jobs = fw.supervised_fine_tuning_jobs.list()
rft_jobs = fw.reinforcement_fine_tuning_jobs.list()
dpo_jobs = fw.dpo_jobs.list()

​
Operational guidance

  • Use managed jobs when your loss/objective fits supported product flows and you do not need per-step custom loss code.
  • If you need custom losses such as GRPO variants or hybrid objectives, switch to service-mode loops (reinforcement_fine_tuning_steps) with local scripts. Service-mode trainer jobs currently support full-parameter tuning only (lora_rank=0).