AdministrationCompute

Compute

With flexible compute, you can define compute resources from Kubernetes clusters that you can then assign to run your generator training and synthetic data generation workloads. You can use compute resources to add GPU-capable compute that can significantly speed up the time to train new generators and generate synthetic datasets.

Add a new Kubernetes compute

You can define any of your existing Kubernetes compute resources as a new compute in MOSTLY AI.

Prerequisites

  • Before you define a new compute, check the available compute resources in your Kubernetes cluster.
  • You need to be a super admin to add a new compute.

Steps

  1. From the profile menu, select Compute. MOSTLY AI - Profile menu - Select Compute
  2. Click + New compute. MOSTLY AI - Click New Compute
  3. Select Kubernetes. MOSTLY AI - Select Kubernetes
  4. Define the compute configuration.
    Kubernetes compute parameterDescription
    NameA unique name for the compute.
    CPUsThe number of CPU cores.
    MemoryThe amount of memory.
    GPUsThe number of GPUs.
    GPU memoryThe amount of GPU memory.
    TolerationA toleration that matches the taint defined for your cluster.

    ⚠️ Leave empty if in the values.yaml you need to set the mostly_coordinator.deployment.core_job.tolerationOperator to Exists.
    Order IndexThe order in which the compute appears in the Compute drop-down.
    The default compute has order_index=1.
    MOSTLY AI - New Kubernetes compute
  5. Click Save.

Result

The next time users configure a new generator training or synthetic dataset, they can select the new compute from the Compute drop-down list.

MOSTLY AI - Result - Select Compute

In addition, the new compute remains available in the Computes list.

MOSTLY AI - New Compute available

Edit a compute

You can make changes to exising computes.

Steps

  1. Open the Compute kebab menu and select Edit.
  2. Change the compute configuration.
  3. Click Save.

Result

The compute configuration is now updated.

Delete a compute

You can delete an existing compute from its kebab menu.

💡

Do not delete the Default compute as all common steps for generators and synthetic datasets rely on its availability.

Steps

  1. Open the Compute kebab menu and select Delete.
  2. Click Yes, delete compute in the confirmation pop-up.