platforms.yaml

# Inference platforms — third-party places to self-host vLLM.
#
# Phase 1: 2 curated platforms. Each model opts in via `meta.platforms: [id, ...]`
# in its YAML; recipes that don't opt in don't render a Deploy button.
#
# Schema:
#   id      — slug referenced by model YAML
#   name    — display name in the dialog
#   url     — "Open guide" external link
#   blurb   — one-line description shown under the name
#   install — (optional) short shell/Dockerfile snippet shown inline. Omit
#             for platforms that don't need any local setup.

platforms:
  - id: modal
    name: Modal
    url: https://modal.com/docs/examples/vllm_inference
    blurb: Serverless Python — wrap vLLM in @app.cls and ship with `modal deploy`
    install: |
      pip install modal
      modal setup

  - id: sagemaker
    name: AWS SageMaker
    url: https://docs.vllm.ai/en/latest/examples/deployment/sagemaker-entrypoint/
    blurb: Managed container — BYO Docker image with the SageMaker entrypoint
    install: |
      # Dockerfile
      FROM vllm/vllm-openai:latest
      COPY sagemaker-entrypoint.sh /app/
      RUN chmod +x /app/sagemaker-entrypoint.sh
      ENTRYPOINT ["/app/sagemaker-entrypoint.sh"]