WangKaiLin/PipeOwl-1.2

PipeOwl-1.2(Geometric Embedding)

A transformer-free semantic retrieval engine.

PipeOwl performs deterministic vocabulary scoring over a static embedding field:

score = α⋅base + β⋅Δfield

where:

• —base = cosine similarity in embedding space
• —Δfield = static scalar field bias

Features:

• —O(n) over vocabulary.
• —No attention.
• —No transformer weights.

Patch Note

1.1

• —fix OOV
• —symbolic fallback
• —english fallback
• —japanese fallback
• —PipeOwlConfig improvement
• —Tokenizer: max_len cap
• —load_assets: contiguous + row-normalize
• —small benchmark

1.2

• —safetensors support

Architecture

• —Static embedding table (V × D)
• —Aligned vocabulary index
• —Optional scalar bias field
• —Linear scoring
• —Pluggable decoder stage
• —Targeted for CPU environments and low-latency systems (e.g. IME).
• —Single static field (~635MB), no runtime model weights.

Attribution

The base embedding vectors were generated using BGE-M3 (Apache-2.0) via inference. This repository does not redistribute any original BGE weights.

Quickstart

pip install numpy safetensors
python quickstart.py

See full experimental notes here:

https://hackmd.io/@galaxy4552/SJ5DatsuZx

Repository Structure

pipeowl1.2/
 ├ README.md
 ├ config.json
 ├ LICENSE
 ├ quickstart.py
 ├ pipeowl.safetensors
 ├ vocabulary.json
 └ engine.py

PipeOwl 是一個基於靜態語義場的幾何檢索系統。

核心公式：

score = α⋅base + β⋅Δfield

其中：

• —base = embedding cosine similarity
• —delta = 靜態場偏移量
• —α / β 為可調權重

提供一種 O(n) 的輕量語義計分方法， 適合低延遲環境（如輸入法）。

LICENSE

MIT