Bring Your Own Open Weights. Tune & Serve Serverlessly on Vertex AI Model Garden.

import vertexai
from vertexai.preview import model_garden

PROJECT_ID = "<YOUR_PROJECT_ID>"
REGION = "us-central1"
BUCKET_URI = f"gs://{PROJECT_ID}-mg-byom-demo"

vertexai.init(project=PROJECT_ID, location=REGION)
custom_model = model_garden.CustomModel(
    gcs_uri=f"{BUCKET_URI}/hf-weights/Qwen--Qwen3-4B"
)

# Query live regional GPU stock + project quota (avoids scarce SKU waits)
options = custom_model.list_deploy_options(
    available_machines=True,
    filter_by_user_quota=True,
)
print(options)

# All-or-none hardware rule: pass machine_type + accelerator_type + accelerator_count together
endpoint = custom_model.deploy(
    machine_type="g4-standard-48",
    accelerator_type="NVIDIA_RTX_PRO_6000",
    accelerator_count=1,
    model_display_name="qwen3-4b-byom-base",
    endpoint_display_name="qwen3-4b-byom-base-ep",
)

# === Dry-Run Output (--emit-iam): Review & Execute in Google Cloud Shell ===
export PROJECT_ID="<YOUR_PROJECT_ID>"
export REGION="us-central1"
export BUCKET_URI="gs://${PROJECT_ID}-mg-byom-demo"
export PROJECT_NUMBER="$(gcloud projects describe "${PROJECT_ID}" --format='value(projectNumber)')"

# 1. Enable required Google Cloud APIs
gcloud services enable \
    aiplatform.googleapis.com \
    compute.googleapis.com \
    storage.googleapis.com \
    logging.googleapis.com \
    --project="${PROJECT_ID}"

# 2. Create regional Cloud Storage bucket with uniform bucket-level access
gcloud storage buckets create "${BUCKET_URI}" \
    --project="${PROJECT_ID}" \
    --location="${REGION}" \
    --uniform-bucket-level-access

# 3. Provision Vertex AI Service Identity
gcloud beta services identity create \
    --service=aiplatform.googleapis.com \
    --project="${PROJECT_ID}"

# 4. Grant least-privilege bucket-scoped Storage Object User (roles/storage.objectUser)
#    to the Vertex AI Service Agent and Managed OSS Fine-Tuning Service Agent
gcloud storage buckets add-iam-policy-binding "${BUCKET_URI}" \
    --member="serviceAccount:service-${PROJECT_NUMBER}@gcp-sa-aiplatform.iam.gserviceaccount.com" \
    --role="roles/storage.objectUser"

gcloud storage buckets add-iam-policy-binding "${BUCKET_URI}" \
    --member="serviceAccount:service-${PROJECT_NUMBER}@gcp-sa-vertex-moss-ft.iam.gserviceaccount.com" \
    --role="roles/storage.objectUser"