Train with Model Factory
Our forward deployed researchers help you choose a model and training method, prepare data and run experiments. We develop or source training environments as needed. You don’t need a finished model or dataset to start.
Model
training
Adapt a multimodal model to your problem with our forward deployed researchers, or run your own training code in Hedra GPU Sandboxes.
Training options
Run your code in Hedra GPU Sandboxes
Run training, rollouts and environments on isolated GPU compute. You control the runtime.
Training methods
Supervised fine-tuning
Adapt a supported model using examples of the inputs it will receive and the outputs you want.
LoRA adapters
Train a lightweight adapter for a new use case while keeping the base model’s weights fixed.
Reinforcement learning
Use reward feedback from rollouts to train the model toward the behavior you need in your environment.
Compare checkpoints
We use held-out cases to check whether training improved the target behavior and preserved the capabilities you rely on.
Prepare for production.
Plan serving around the selected model, hardware and performance targets. Agree checkpoint handoff, deployment responsibilities and terms with our team, then use production evidence to guide the next training run.
