A standalone Next.js app that turns maintained FastVideo inference recipes into executable YAML or Python. The docs build exports it beneath the GitHub Pages base path at config-generator/.

Two pages are available:

Quick Start (/config-generator/) — choose a task, model profile, and GPU count.
Advanced Tuning (/config-generator/tuning/) — start from an exact model recipe and adjust supported configuration fields.

Run locally

pnpm install
pnpm dev

Open http://localhost:3003/config-generator/.

Build

pnpm build

The build first runs scripts/validate-data.mjs, then writes a static export to out/.

Recipe data

data/tuning.json is the app's recipe source. Its source block pins the exact FastVideo revision and repository paths used to transcribe model defaults, pipeline settings, examples, and verified attention compatibility.

data/quickstart.json contains only the curated task/profile choices and engine offload settings. Each profile references a model from tuning.json; the build fails on missing models, task mismatches, duplicate profiles, unknown attention backends, or unsupported performance fields.

The guide intentionally publishes no latency, VRAM, RAM, or speedup claims. Add those only when a reproducible benchmark record includes the FastVideo and model revisions, GPU type/count, complete resolved recipe, software environment, prompt/seed, warmup/sample protocol, latency statistic, and peak memory.

Updating a recipe

Check the current registry, model-family preset, pipeline config, maintained example, and support matrix.
Update the model entry in data/tuning.json and advance source.fastvideoCommit to the inspected revision.
Update data/quickstart.json only if the curated profile selection changes.
Run pnpm build.

The UI reads the JSON directly; model defaults do not belong in the .tsx components.