Skip to main content
You can tune models for free on Fireworks. Models under 16B parameters are available for free tuning—when creating a fine-tuning job in the UI, filter for free tuning models in the model selection area on the fine-tuning creation page. If kicking off jobs from the terminal, you can find the model ID from the Model Library.
Fireworks RFT helps you train frontier models like DeepSeek V3 and Kimi K2 to outperform closed models for your product use case, using reinforcement learning. Fireworks RFT is powerful and easy to use for developers and enterprises:
  • No infrastructure: Train frontier models without managing GPUs or RL infra
  • Production-ready: Built-in tracing, monitoring, security & one-click deploy
  • Fast iteration: From evaluator setup to deployed model in hours, not weeks
See how Genspark and Vercel used Fireworks RFT to train open models for agentic use cases, outperforming leading closed models.

​
Quickstart: Pick Your Training Approach

Single-Turn Training

⏱️ 15 minutesBest for: Testing locally, simple task trainingHow it works: Iterate on your evaluator and use it to train a small model on Fireworks.

Remote Agents

⏱️ 1-2 hoursBest for: Agents, multi-turn workflows, existing servicesHow it works: Rollouts happen in your environment. Connect via HTTP with tracing.

Secure Training (BYOB)

⏱️ 2-4 hoursBest for: Sensitive data, compliance, enterpriseHow it works: Training data never leaves your GCS/S3 bucket. Full data isolation.

Training SDK (Tinker compatible)

Best for: Expert teams needing full-parameter updates and custom RL objectivesHow it works: Connect to service-mode trainers and run custom training loops with Tinker-compatible APIs.Availability: Full-parameter RFT is currently in private preview and only supported through Training SDK workflows. In this service-mode path, LoRA tuning is not yet supported (lora_rank/loraRank must be 0). Use the Full Parameter RL Tuning guide to request access and get started.
Looking for the private-preview full-parameter path? Start here: Tinker API Compatibility & Full Parameter Tuning.

​
Differentiation and flexibility

Fireworks supports both managed training products and custom Training SDK loops on the same platform. Teams can start with managed flows for fast baselines, then move to custom loops when research requirements outgrow standard objectives.
ModeWhen to useObjective/loss supportStarter path
Managed SFTYou want fastest supervised adaptation with minimal custom codeManaged supervised fine-tuning objective/api-reference/training-sdk/sft-example
Managed DPOYou have preference pairs and want a managed preference-learning flowManaged DPO objective/api-reference/training-sdk/managed-jobs
Managed RFTYou want managed reinforcement fine-tuning with Fireworks orchestrationManaged reinforcement fine-tuning objective/fine-tuning/reinforcement-fine-tuning-models
Training SDK (Tinker compatible)You need maximum flexibility, full-parameter control, or custom objectivesCustom losses/objectives in local loop code (for example GRPO, DPO variants, hybrid RL losses). Full-parameter RFT is currently available only through Training SDK (private preview)./api-reference/training-sdk/overview
Local starter scripts for customer-local implementation:

​
Launch Training

Prerequisites & Validation

Requirements, validation checks, and common errors before launching

CLI (eval-protocol)

Fast, scriptable, reproducible. Perfect for automation and iteration

Web UI

Visual, guided, beginner-friendly. Great for exploring options
Already familiar with firectl? You can create RFT jobs directly.

​
RFT Concepts

How RFT Works

The RL training loop explained

Evaluators

How reward functions guide training

Environments

Local vs remote evaluation environments

Parameter Tuning

Optimize your training configuration

Cost Estimator

Estimate and optimize your training costs