Prebuilt Chroma reference index for slugline-mcp: the full rohitsaxena/MovieSum dataset (~2,200 screenplays, 288,091 scenes), embedded with sentence-transformers/all-MiniLM-L6-v2 and mood-tagged with a full 25-label zero-shot classifier (facebook/bart-large-mnli, batched, no shortlist).
This tag is deliberately separate from the package's own version tags (e.g. v0.1.0) — the reference index can be rebuilt and republished under a new data-vN tag independently of any PyPI release.
Asset: full-index.zip (~1.8GB) — a zipped Chroma persistent-client directory (chroma.sqlite3 + HNSW index folder). slugline-mcp's bootstrap module downloads and unpacks this automatically on first run if no local index exists.
Built on a rented GPU; see remote-index/DEPLOY.md in this repo for the reproduction steps.