Skip to content

feat: /ml catalogue generated from models.yaml (35 models) - #221

Merged
ronaldtse merged 2 commits into
mainfrom
feat/generated-model-catalog
Nov 1, 2026
Merged

ronaldtse merged 2 commits into
mainfrom
feat/generated-model-catalog

Conversation

@ronaldtse

Copy link
Copy Markdown
Contributor

Summary

The /ml page was hand-written and stale (14 of the current 35 models; "index v8 resolves 26"). Now the ledger generates at build time from interscript-models/models.yaml — npm run generate:models — with per-model id, task, precision, size, split-part count, metrics in priority order, and a link to its exact GitHub release. Tiers split on precision (fp32 server / quantized client); prose counts compute from the data. Also updates the maps-page test's hardcoded 289 to the true 318.

Test plan

  • astro check: 0 errors
  • vitest: 239 passed; the only failure (api-endpoints dev-server startup timeout) reproduces identically on a clean main worktree — pre-existing environment flake, not this diff
  • /ml probed on the dev server: 200, "resolves 35 models", renamed ids render

/docs/libraries 404s live — the Libraries doc generates at
/docs/Libraries (loader preserves filename stems). Both hand-written
hrefs updated; targets verified against production.
The page was hand-written and listed 14 of what is now 35 models.
scripts/generate-models.mjs reads the registry (same pattern as the
maps catalogue generator): id, task, precision, size, split-part count,
sha-pinned release link, and ordered metrics per model; ml.astro
renders the ledger from that data, split by tier on precision, with
every id linking to its release. Prose counts now compute. The maps
page test's hardcoded 289 became 318 (the true current total).
@ronaldtse
ronaldtse merged commit ef0418d into main Nov 1, 2026
2 of 3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant