Custom inference
Run your own inference pipeline with custom model code and kernels.
Access by request
Hedra GPU
Sandboxes
Bring your container, dependencies and model code. Run custom inference, training and evaluation in an isolated GPU environment.
Your model code · PyTorch
import torch
model = model.to("cuda").eval()
inputs = inputs.to("cuda")
with torch.inference_mode():
outputs = model(inputs)
What you can run
Reinforcement learning
Run training loops and rollouts in your own environments.
Synthetic data
Generate labeled data from simulated scenes, desktops or sensor feeds.
Model evaluation
Run models on the same inputs and compare their scores.
You control the runtime.
Hedra runs the infrastructure.
You control
Container image and dependencies
Model code, custom kernels and inference pipeline
Training loops, environments, rollouts and evaluations
GPU requirements for each job
Hedra manages
Job placement on supported NVIDIA and AMD GPUs
Infrastructure scaling
Recovery from hardware failures
Capacity across our data center partners
From request to first job.
Describe your workload.
Tell us what you’re running, the GPUs you need, your storage and networking requirements, and when you need to start.
We confirm capacity and terms.
Capacity is limited. We confirm the supported configuration, allocation, job limits, pricing and start date. A request does not reserve capacity.
Run your first job.
Once we confirm access, set up your environment and run the job. We’ll help you validate the outputs.
