What problem does it solve? Deploying machine-learning demos on Hugging Face Spaces involves many platform-specific pitfalls: ZeroGPU runtime constraints, hardware selection, dependency pinning, silent build failures, and stateless storage. This Skill guides an agent through the full lifecycle of creating, building, verifying, and debugging a Space so it actually works in production. ## Core Features & Use Cases - Space creation and configuration: Create Gradio, Docker, or Static Spaces with the correct SDK, hardware flavor (cpu-basic, ZeroGPU, dedicated GPU), secrets, and README frontmatter via the hf CLI. - ZeroGPU deployment patterns: Enforces the critical rules for ZeroGPU apps, such as importing spaces before torch, loading models at module scope, and decorating functions with @spaces.GPU. - Verification and debugging workflows: Provides a four-step smoke-test procedure (stage check, log inspection, gradio_client API call, output validation) plus a rung ladder from hot-reload to factory reboot, and a catalog of known errors. - Use Case: A user asks to deploy a finetuned Hunyuan3D model as a public demo. The Skill checks for prior art, duplicates the official Space, swaps the checkpoint repo, pushes with hf upload --repo-type space, and verifies the live endpoint returns a valid GLB mesh. ## Quick Start Ask the agent to create and deploy a Hugging Face Space for your model, for example: "Build a Gradio Space on ZeroGPU that runs my diffusion model and verify the API works."
