Open the ROCm Readiness dashboard
A data pipeline for ROCm engine qualification and model performance. GitHub Actions
collects producer feeds and publishes JSON under main/data/. It includes stable
ROCm llama.cpp builds and their benchmark results.
The dashboard has two views of that dataset. ROCm readiness shows each engine's builds and model performance for a ROCm version, with stable and nightly channels. Upstream health shows Radeon and Ryzen AI CI in each project's own repository on its default branch, configured in sources.json. See Upstream health.
Requires uv and an authenticated GitHub CLI (gh) for benchmark package evidence.
uv run --no-project --python 3.12 python scripts/check.py --output-dir data/local-previewThis runs the tests and collection used by CI. Readiness discovers published
ROCm binaries from GitHub releases and matches qualification reports to the
selected archives. Set GH_TOKEN for GitHub API access. UPSTREAM_API_BUDGET
caps calls for the upstream health view. Add --previous-data data/local-preview
with a different output directory to reuse the release inventory. PR checks
use --fresh to rebuild from source. The preview includes source copies and
diagnostics for inspection.
Actions runs on PRs, main pushes, manual dispatch, and a daily schedule. Its
readiness-data-* artifacts are retained for 30 days. Main runs publish successful
data; failed runs preserve the last good pointer and report errors in sync status.
- Data flow: ingestion and published-file diagrams.
- Data pipeline: source setup, adapters, channels, and package mappings.
- Upstream health: sources.json, collection rules, the upstream badge and local runs.
- Data publication: runtime files and ROCm directory names.
- Build-issue notifications: post-collection build alerts, delivery state, and safe setup.
- JSON reference: field contracts, source mappings, and validation.
- Dashboard UI: local browser preview, static build and UI checks.
- Signal Deck design language governs dashboard UI changes.