poolin

A local UI package for pooling existing embedding vectors from an embedding zip into grouped higher-level vectors.

Important note

Standard sentence-transformer style pooling usually happens inside the embedding model when token embeddings are converted into one sentence embedding. This package does post-embedding vector pooling over already-created chunk embeddings.

What it does

• launches with the poolin command
• reads an embedding zip such as RAG_chunks_recursive_chunks_embeddings.zip
• auto-groups related chunk embeddings by filename pattern like RAG_chunk_001_rcs_001.md -> RAG_chunk_001
• pools vectors with one of these methods:
  • auto
  • mean
  • max
  • weighted_char_mean
  • weighted_word_mean
  • mean_sqrt_len
• exports a zip with:
  • pooling_summary.json
  • pooling_manifest.csv
  • *_pooled_embeddings.jsonl (optional)
  • *_pooled_embeddings.csv (optional)
  • *_pooled_embeddings.npz (optional)

Install

pip install poolin

Run

poolin

Suggested input

Use a zip produced by your embedding step, containing an embeddings .npz or .jsonl payload plus the summary file.

Ownership note

The package metadata and copyright notice are set to Wenxi Wang. You should still verify PyPI package-name availability, trademark questions, and any legal or patent issues yourself before publishing.