Source:    https://huggingface.co/minishlab/potion-base-8M
Revision:  bf8b056651a2c21b8d2565580b8569da283cab23
License:   MIT
Copyright: (c) MinishLab

Vendored verbatim into kaos-nlp-transformers on 2026-05-08 for offline
loading via the model2vec backend (see kaos_nlp_transformers.embedding
._load_model2vec_cached). The on-disk bytes match the upstream HF Hub
snapshot at the pinned revision; no modifications.

The kl3m / minishlab attribution chain documents the model's provenance:
distilled from BAAI/bge-base-en-v1.5 via the Tokenlearn + POTION recipe
with PCA reduction to 256 dim. See
https://minishlab.github.io/tokenlearn_blogpost/ for the methodology.

The MIT license terms apply to the safetensors weights, the tokenizer,
and the included config files. No re-licensing — kaos-nlp-transformers
is Apache-2.0, but vendored third-party artifacts retain their upstream
licenses.

Files in this directory:
  config.json                          (~190 B)
  modules.json                         (~150 B)
  special_tokens_map.json              (~125 B)
  tokenizer_config.json                (~1 KB)
  tokenizer.json                       (~700 KB)
  vocab.txt                            (~225 KB)
  model.safetensors                    (~30 MB — the embedding matrix)
  README.md                            (upstream's README, ~10 KB)
