PipeOwl-1.8.2-jp-Temperature (Geometric Embedding)

A transformer-free semantic retrieval engine.

PipeOwl performs deterministic vocabulary scoring over a static embedding field:

score = α⋅base + (1 - α⋅base)⋅Δfield

• BPB：用 byte 當單位
• token NLL：用 token 當單位

token NLL: 12.943284891453972

where:

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

Features:

• O(n) over vocabulary.
• No attention.
• No transformer weights.
• CPU-friendly (<16MB model)

Architecture

• Static embedding table (V × D)
• Aligned vocabulary index
• Optional scalar bias field (Δfield)
• Linear scoring
• Pluggable decoder stage
• Targeted for CPU environments and low-latency systems (e.g. IME).

Model Specs

item	value
vocab size	26155
embedding dim	256
storage format	safetensors (FP16)
model size	~13.2 MB
languages	Japanese
startup time	<1s
query latency	34 ms (CPU, full vocabulary scan)

Quickstart

git clone https://huggingface.co/WangKaiLin/PipeOwl-1.8.2-jp-Temperature
cd PipeOwl-1.8.2-jp-Temperature

pip install numpy safetensors

python quickstart.py

Example:

Example semantic retrieval results:

Please enter words： 東京

Top-K Tokens:
1.000 | 東京
0.880 | は
0.790 | 大阪
0.766 | パリ
0.747 | 名古屋

Please enter words： 大阪

Top-K Tokens:
1.000 | 大阪
0.889 | は
0.832 | 東京
0.817 | 関西
0.816 | 尼崎

Repository Structure

PipeOwl-1.8.2-jp-Temperature/
 ├ README.md
 ├ config.json
 ├ DATA_SOURCES.md
 ├ eval_bpb.py
 ├ parameter.txt
 ├ LICENSE
 ├ quickstart.py
 ├ engine.py
 ├ vocabulary.json
 └ pipeowl_fp16.safetensors

LICENSE

MIT