Skip to main content
Fireworks helps you fine-tune models to improve quality and performance for your product use cases, without the burden of building & maintaining your own training infrastructure.
Coming from OpenAI? Fireworks uses the same OpenAI-compatible chat completion format for training data — the same messages array with role, content, tool_calls, and weight fields. You can use your existing SFT datasets with no conversion required. See our OpenAI compatibility guide for more details.

​
Before managed or dedicated training: account tier and GPU quota

Managed jobs and dedicated Training API runs need training GPU quota, granted automatically by spending tier: Check your quota with firectl quota list. A job rejected with HTTP 429 quota_exceeded (sometimes a 403 on the job poll) is a tier issue, not a dataset/config problem. Serverless Training uses a shared pool with its own model, concurrency, and rate limits instead of dedicated training GPU quota.
Need more training quota than your tier allows? Reach out for enterprise support and we’ll help size the right allocation for your workload.

​
Start here

Use Choose a Training Path to choose managed or Training API, serverless or dedicated infrastructure, and the interface you want to use.

Choose a training path

Compare workflows, infrastructure, and interfaces.

Train from your coding agent

Install one skill and ask in plain language.

​
When to use SFT vs. RFT

In supervised fine-tuning, you provide a dataset with labeled examples of “good” outputs. In reinforcement fine-tuning, you provide a grader function that can be used to score the model’s outputs. The model is iteratively trained to produce outputs that maximize this score. Supervised fine-tuning (SFT) works well for many common scenarios, especially when:
  • You have a sizable dataset (~1000+ examples) with high-quality, ground-truth labels.
  • The dataset covers most possible input scenarios.
  • Tasks are relatively straightforward, such as:
    • Classification
    • Content extraction
However, SFT may struggle in situations where:
  • Your dataset is small.
  • You lack ground-truth outputs (a.k.a. “golden generations”).
  • The task requires multi-step reasoning.
Here is a simple decision tree:
Verifiable refers to whether it is relatively easy to make a judgement on the quality of the model generation.
For custom losses, rollouts, per-step control, or algorithm research, continue to the Training API overview.