Skip to content

codewhale-lmm-provider v0.1.0-alpha.3

Pre-release
Pre-release

Choose a tag to compare

@LIghtJUNction LIghtJUNction released this 17 Nov 10:22
2b0d6b2

Model catalog and metadata repair

  • Discover models from the authorized account catalog without a bundled model-name allowlist.
  • Fetch public metadata from models.dev with a 24-hour automatic disk cache, ETag revalidation, shared in-flight refresh, and stale-cache fallback.
  • Match names tolerantly across provider prefixes, case, separators, aliases, and dates while retaining exact account routing identifiers. Ambiguous or unknown metadata does not remove supported models.
  • Preserve account limits and prices; treat zero or missing public capacities as unknown instead of dropping the route. Keep unsupported specialized models separate from chat models.
  • Accept large model catalogs; remove obsolete bundled model snapshots.

Validation

44 existing tests, 19 focused adapter checks, and the 5 MB catalog regression passed.
All validation used synthetic credentials and local fixtures; no production paid model call was made.

Source: 2b0d6b2
Artifact SHA-256: 54b3ecaa85ce1a6de3573a4e14b030eda45338c2312fc7a195b8bdaefcac8da4

The attached artifact is the verified installation package. This release does not imply npm or Visual Studio Marketplace publication.