Guides
How to fine-tune something that works.
Where should I start?
If you are not sure fine-tuning is the right answer, start with whether it is a prompt problem. If you have decided and need to size a run, start with how much VRAM you need. If a fine-tune has already disappointed you, start with why fine-tuned models fail.
What these are for
Fine-tuning has an unusually wide gap between the tutorial and the practice. The tutorial trains a model on a public dataset and reports a loss curve. The practice involves deciding whether the project is worth doing at all, discovering that a fifth of your examples are silently truncated, and explaining to somebody why the model that scored well is producing worse answers than the one it replaced.
These guides are about the second thing. Each opens with a direct answer to the question in its title, and each is written to be useful whether or not you use this platform — the memory arithmetic, the failure modes and the evaluation design are properties of the problem, not of us.
Where our own product is the answer, the guide says so and links to it. Where it is not — custom training loops, multi-node distribution, running on your own hardware — the guide says that instead and names what does solve it.
Getting started
Should you fine-tune, or is it a prompt problem?
When is fine-tuning worth it compared to prompt engineering or retrieval?
How to choose a base model
How do I choose which base model to fine-tune?
Migrating from a closed model API to an open-weight fine-tune
How do I replace a closed model API with a fine-tuned open-weight model?
Preparing data
Training
LoRA, QLoRA or full fine-tuning: how to choose
What is the difference between LoRA, QLoRA and full fine-tuning?
Why fine-tuned models fail, and how to tell which failure you have
Why did my fine-tuned model get worse instead of better?
When to use preference tuning instead of supervised fine-tuning
When should I use DPO instead of supervised fine-tuning?
Using reinforcement learning to improve reasoning
How does GRPO work and when should I use reinforcement learning to fine-tune?
Evaluation
Serving
Scale to zero or always warm: choosing a serving mode
Should my model endpoint scale to zero or stay always warm?
Serving many fine-tuned models without paying for each one
How can I serve dozens of fine-tuned models economically?
Keeping a fine-tuned model portable
How do I make sure I can move a fine-tuned model to another provider?
Cost
How much VRAM does fine-tuning actually need?
How much GPU memory do I need to fine-tune a model?
How to estimate what a fine-tuning run will cost
How much does it cost to fine-tune an open-weight model?
Controlling spend on a fine-tuning platform
How do I stop fine-tuning and inference costs running away?
Start with the free tier
A magic link creates your account, your tenant and your first API key. No card until you ask for compute.