Create a training job

Pick a framework, configure resources, mount storage and image — submit a training job from the HyperTrain console

HyperTrain is a Kubernetes-native distributed training service with built-in PyTorch, DeepSpeed, MPI, and TensorFlow frameworks. The platform abstracts infra, scheduling, and runtime dependencies into a unified service interface — so users can launch training jobs without managing the underlying ops.

Prerequisites

Compute account DCU balance > 0
Cash account balance > 0
The enterprise has provisioned NAS bulk or NAS performance storage in the current data center, with permissions on the current account.
For private images, the enterprise must have an image registry in the same data center.

Steps

Sign in and go to Product Center → Compute → HyperTrain. Click Activate or Create Job to enter the job creation page.

Job creation entry

1. Basic information

Basic information

2. Resource configuration

Resource configuration

3. Storage and image

Storage and image

4. Other settings

Other settings

Submit the job. The job appears in the list page once successfully created.

Field reference

Field	Description
Job name	Auto-generated by default; customizable.
Template	Create from an existing template, or skip.
Description	Free-form summary of the job.
Region	Data center where the job runs.
Framework	PyTorch, DeepSpeed, MPI, TensorFlow.
Resources	Compute spec and node count.
Storage	Storage type and mount path.
Image	Choose from base image, app image, or private image. Private = your custom image stored in the enterprise registry.
Env vars (optional)	Custom env. The platform also injects system variables automatically.
Auto-retry	Retry the job up to N times on failure.
Timeout	Hard cap on wall-clock runtime; the job auto-cancels on timeout.
Start command	Default working dir for platform images is `/root`; for custom images the dir set in the image is used.
Priority	Applies only to queued jobs. Lower number = higher priority.