Start with a supported model
Choose a Hedra or open model, or bring your own supported weights.
Build a custom model
Start with your own model or an open-weight one. Hedra provides the infrastructure and expertise to fine-tune, evaluate, and serve custom models.
From training to serving.
Train and evaluate
Run your training code and compare checkpoints on your task. We handle distributed execution.
Serve with WMR
Deploy your checkpoint with WMR. Use production feedback to guide the next training run.
Train with your data and rewards.
Supervised fine-tuning
Adapt a VLA or open model to your embodiment, domain, or sensor data.
LoRA
Train lightweight adapters for task-specific variants of one base model.
Reinforcement learning
Train against your reward signal, with generation and training on the same GPU.
Model evaluation
Compare checkpoints on the same task before deploying with WMR.
Your data. Your stack.
Keep your data private.
Your data is not used to train Hedra’s models. Deploy on Hedra or export your checkpoints.
Run your own stack.
Use GPU Sandboxes for custom runtimes, environments, and rollout loops.
Bring us your model and task.
Tell us what your model needs to do. We’ll help you plan training and deployment.
