feat: implement generator-based retrieval to improve memory efficiency #1
Conversation
TecSong
commented
Feb 3, 2026
TecSong
commented
- Add retrieve_iterative method that uses generators to reduce memory footprint
- Implement iterative retrieval with optional batching
- Add answer_stream_iterative method that leverages the new retrieval approach
- Maintain backward compatibility with existing methods
- Add retrieve_iterative method that uses generators to reduce memory footprint - Implement iterative retrieval with optional batching - Add answer_stream_iterative method that leverages the new retrieval approach - Maintain backward compatibility with existing methods
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: a864e9f95c
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| fused_score = ( | ||
| settings.vector_weight * vector_norm.get(chunk_id, 0.0) | ||
| + settings.bm25_weight * bm2_norm.get(chunk_id, 0.0) |
There was a problem hiding this comment.
Calling retrieve_iterative will raise a NameError because the fusion score uses bm2_norm, which is never defined in this method. This means any caller of the new iterative retrieval path will crash before yielding results. It looks like a typo for bm25_norm, which is what the non-iterative path uses for BM25 normalization.
Useful? React with 👍 / 👎.
1 participant
