deployment/deploy_endpoint.py

from huggingface_hub import create_inference_endpoint

def deploy():
    endpoint = create_inference_endpoint(
        name="ericsson-day-demo-endpoint",
        repository="ED-Demo/Ericsson_day_demo_model",
        task="text-classification",
        framework="pytorch",          # 👈 required
        vendor="aws",
        region="eu-west-1",
        accelerator="cpu",
        instance_size="small",        # size (small, medium, large)
        instance_type="intel-icl",    # 👈 required, CPU family (Intel Ice Lake)
        type="protected",             # requires HF token
    ).wait()

    print("✅ Endpoint deployed at:", endpoint.url)

if __name__ == "__main__":
    deploy()