Applications / Template

Model / GPU

GitHub example manifest: applications/model/gpu.yaml

Overview

This example is available in the AstroPulse GitHub template repository. Review its manifest and deployment requirements before applying it.

Category
Applications
Source
applications/model/gpu.yaml

This manifest was discovered from GitHub. Detailed prerequisites and commands have not been curated yet; do not assume it is ready for direct deployment.

Prerequisites & dependencies

Prerequisites have not been documented for this GitHub example. Inspect the manifest before use.

No dependencies are listed in the available catalog metadata. This is not a guarantee that none are required.

Source manifest

applications/model/gpu.yaml

# A Hugging Face model on one GPU, scaling with demand
#
# Qwen3 8B on an NVIDIA L4 per replica. Replicas scale from 1 to 4 on the
# requests waiting for each replica.
#
# Before you apply:
#   - a profile (profiles/)
#   - GPU capacity: a GPU node group (clusters/aws/eks/gpu-node-group.yaml) or
#     the karpenter GPU node pool (add-ons/karpenter-gpu-pool.yaml)
#   - the cert-manager, model-serving, cost-visibility and
#     workload-autoscaling add-ons (add-ons/)
#   - for a gated model, connect Hugging Face once for your organization:
#     echo "$HF_TOKEN" | astroctl integrations huggingface connect --token-stdin
#
# Apply:  astroctl app apply -f gpu.yaml --dry-run, then without --dry-run
apiVersion: platform.astropulse.io/v1
kind: Application
spec:
  name: qwen-chat
  profileName: <YOUR_PROFILE_NAME>
  source:
    type: Model
    model:
      uri: hf://Qwen/Qwen3-8B          # add :<revision> to pin a commit
      serving:
        apiFormat: OpenAI
        maxContextTokens: 8192
      resources:
        accelerator:
          gpu:
            model: l4
            countPerReplica: 1
      scaling:
        minReplicas: 1
        maxReplicas: 4
        metric:
          requestsWaiting:
            target: "2"               # requests waiting per replica

Deployment guidance

Go straight to Console to create clusters, deploy applications, and manage supported resources. Templates and CLI commands are optional; not every GitHub manifest has an automatic import path.

Console — start directly

Open Console to create or connect infrastructure and deploy applications without running astroctl or copying this YAML first. This template may still require manual setup.

Open Console

astroctl

No verified CLI sequence is published for this example. Review the source and the onboarding guide rather than running a guessed command.

Get Started

Nova

Prefer to describe the goal? Ask Nova to review this example and help plan supported actions. Access, policy, and approvals still apply.