Fireworks announced that its Training API and Fireworks Lab are generally available. The API lets developers run their own training loops in Python while Fireworks provides the compute and handles model rollout. Users can start instantly with a serverless option that charges per token, or request dedicated GPU capacity for larger runs. The platform claims to cut iteration time by 2–4× and reduce weight‑transfer bandwidth by up to ten times.
Why it matters
Teams can fine‑tune models faster and cheaper without building their own GPU clusters.
