Models

Fine-tune an open-weight model on the data your agent has already produced. You pick the model, Overmind runs the training on GPUs, streams the loss curves while it goes, and scores the finished model against the one your capability runs in production — with the same judges you use everywhere else. The model is yours. Its weights are downloadable, and it serves through an OpenAI-compatible API you call with your existing client.

​

What you need

Your train and eval data must not share the same runs. Overmind counts the overlap for you and shows it in the wizard before you launch; anything overlapping is dropped from training rather than quietly inflating your score.

​

Pick a model

The wizard ranks the catalogue for your data and estimates cost and duration for each option before you spend anything. Three things can rule a model out, and the wizard handles all three for you:

• Context. A model’s training context is often shorter than its inference window. One that can’t fit your longest row is dropped from the recommendations.
• Tool calling. If your data contains tool calls, only models that support them are offered.
• Adapters. Larger dense models and every mixture-of-experts model train as LoRA adapters; the picker says so instead of offering a choice.

​

Launch

One run can launch several experiments at once, so you can try two learning rates or two base models side by side and compare them on the same page. Each experiment has its own Epochs, Learning rate and Batch, plus Rank, Alpha and Dropout when it’s training a LoRA adapter. The panel tells you what your data looks like — rows, tokens, longest row, whether it contains tool calls — so the numbers aren’t guesswork. Your coding agent can do the whole thing: /overmind finetune checks you’re ready, estimates the cost, launches the run, and follows it to the end.

​

Watch it train

A job moves from queued, through training, to deploying, and lands on succeeded. While it runs you get per-step train and validation loss, learning rate, token accuracy and gradient norm, alongside steps completed, an ETA, tokens processed and credits used. You can cancel a run at any point, and retry one that failed.

​

The benchmark

This is the part that tells you whether the training was worth it. Every run with eval data and graders scores two things on the same rows with the same judges:

1. Baseline — the model your capability runs today, captured when you launched.
2. Final — your trained model, once it’s serving.

The headline is the delta between them, with each grader’s score underneath and a link to the full eval run behind every number. When your expected outputs are labels, you also get per-class precision, recall and F1, accuracy, and a confusion matrix — so a classifier tells you which class it’s getting wrong, not just that it improved.

​

Ship it

A finished job deploys its model for you. It appears on Inference and in your capability’s Models tab with a reference like ft-9635f347-qwen3-4b, and its weights are downloadable from the model page or with overmind model download-checkpoint. Two ways to put it in front of traffic:

• Make live points your capability’s alias overmind/<capability-uuid> at the new model. Your next request hits it, with no code change at all.
• Copy gives you a prompt for your coding agent that retargets your LLM client at the model directly. The edit lands in your repository through your normal review.

​

Good to know

• Quality comes from the data, not the hyperparameters. Get the version fitting the train contract first.
• A run freezes the version it trained on, so its numbers stay true while you keep curating for the next one.
• Every experiment reports what it cost when it finishes, and you see the estimate before you launch.
• A model serving real traffic is producing traces again — which is the data for your next round.

​

Next steps