# DATA_SOURCES.md

This document lists the lexical sources used to construct the PipeOwl vocabulary.

Only vocabulary tokens were extracted from these sources.  
No dictionary definitions or explanatory text are included in the model assets.

## Base embedding model

The base embedding vectors were generated using the model:

bert-base-japanese-v3 (Apache-2.0)

During preprocessing, vocabulary tokens were embedded via inference to produce the static embedding table used by PipeOwl.

This repository does **not** redistribute any original bert-base-japanese-v3 model weights.

Model repository:
https://huggingface.co/tohoku-nlp/bert-base-japanese-v3