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.Documentation Index
Fetch the complete documentation index at: https://docs.fireworks.ai/llms.txt
Use this file to discover all available pages before exploring further.
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.Two ways to fine-tune
Fireworks offers two fundamentally different approaches to fine-tuning. Choose the one that fits your needs:Managed Fine-Tuning
Give Fireworks your data and configuration. The platform handles scheduling, training, checkpointing, and model output. No custom code required.Best for teams that want managed SFT, DPO, or RFT with LoRA or full-parameter tuning.
Training API (Tinker compatible)
Write custom Python training loops. You control the loss function, optimizer step, checkpointing, and weight sync. Fireworks handles the distributed GPU infrastructure.Best for research teams needing custom loops, custom rollout orchestration, or inference-in-the-loop evaluation.
| Managed Fine-Tuning | Training API | |
|---|---|---|
| Control | Configuration-driven | Full Python loop control |
| Objectives | Built-in SFT, DPO, RFT | Recipe objectives or any custom loss function |
| Tuning method | Full-parameter or LoRA | Full-parameter or LoRA |
| Inference during training | Not available | Hotload + sample mid-training |
| Interface | UI, firectl, REST API | Python API |
| Best for | Production fine-tuning with standard methods | Research, custom RL, hybrid losses |
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
- Your dataset is small.
- You lack ground-truth outputs (a.k.a. “golden generations”).
- The task requires multi-step reasoning.
When to use the Training API instead
Move from managed fine-tuning to the Training API when you need:- Custom training logic — hybrid objectives, custom reward shaping, or a non-standard algorithm beyond managed settings
- Inference-in-the-loop evaluation — hotload checkpoints onto a serving deployment and sample mid-training
- Per-step control — custom gradient accumulation, dynamic learning rate schedules, or algorithm research
Detailed capability comparison
| Capability | Managed RFT | Training API |
|---|---|---|
| Launch training | CLI or UI | Python script |
| Loss functions | grpo, dapo, gspo-token (built-in) | Any custom loss via forward_backward_custom |
| Tuning modes | Full-parameter or LoRA | Full-parameter or LoRA |
| Context length | Full context length supported by the selected training shape | Full context length supported by the selected training shape |
| Training loop | Fully managed | You write the loop |
| Per-step diagnostics | Dashboard (reward, loss, rollouts) | Full Python access to all metrics |
| Zero-variance filtering | Automatic | You implement |
| Checkpoint management | Automatic | You control via save_weights_for_sampler_ext |
Migrating from managed flow to Training API
If you’ve been using managed RFT and want more control — custom loss functions, richer diagnostics, or algorithm experimentation — the Training API lets you implement your own training loop while keeping the same GPU infrastructure. Managed jobs and cookbook recipes now use the same core tuning capabilities, including LoRA or full-parameter tuning and the full context length supported by the selected training shape.MoE models and Routing Replay
For Mixture-of-Experts (MoE) models like Kimi K2 (384 experts), training stability benefits from Routing Replay — caching the expert routing assignments from the reference policy’s forward pass and replaying them during the training forward pass. This ensures that the same experts process the same tokens in both the reference and policy models, reducing gradient noise from routing changes. Routing Replay is available in the Training API via theloss_fn_inputs mechanism — you can pass routing matrices from the reference forward pass into the training datum. Use the Training API when you need to inspect or customize those forward-pass inputs directly.