Skip to content

feat: harden CLIP image RAG#364

Closed
raoabinav wants to merge 5 commits into
StarTrail-org:mainfrom
raoabinav:feat/clip-image-retrieval
Closed

feat: harden CLIP image RAG#364
raoabinav wants to merge 5 commits into
StarTrail-org:mainfrom
raoabinav:feat/clip-image-retrieval

Conversation

@raoabinav

Copy link
Copy Markdown
Contributor

Evaluator Summary

  • Abi hardened LEANN's CLIP image retrieval path so local image corpora can be indexed with deterministic discovery, stable IDs, rich metadata, and text-to-image search.
  • Design choice: image passages use stable image-<hash> identifiers and store file/path/size/dimension/model metadata, which makes rebuilds reproducible and result inspection reviewer-friendly.
  • The implementation builds from precomputed CLIP embedding arrays and disables query-time recomputation for image vectors, because stored text cannot recreate the original pixels.
  • Quality bar: 6 targeted fake-encoder tests avoid model downloads while covering discovery, metadata, stable IDs, and retrieval wiring, plus lint, format, type, lockfile, and whitespace checks.

@raoabinav

Copy link
Copy Markdown
Contributor Author

Parking this PR for now to keep the review queue focused on #347. I’ll revisit/reopen this work later as a smaller clean PR once the content-hash PR lands.

@raoabinav raoabinav closed this Jun 6, 2026
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