# Third-party notices — emo

The `emo` model derives from components licensed by third parties. Their licenses
apply to the components named below; nothing in the Desert Ant Labs
Source-Available License overrides them.

## Semantic embedding

### potion-multilingual-128M — MinishLab
- **Source:** [`minishlab/potion-multilingual-128M`](https://huggingface.co/minishlab/potion-multilingual-128M)
- **License:** MIT
- **Use in `emo`:** the semantic stream. The shipped semantic table is a
  PCA-reduced (112-dim) and vocab-pruned (~45k token) derivative of this model's
  embeddings. Its SentencePiece tokenizer (XLM-RoBERTa lineage) is used at
  inference to tokenize text for that stream.

### bge-m3 — BAAI
- **Source:** [`BAAI/bge-m3`](https://huggingface.co/BAAI/bge-m3)
- **License:** MIT
- **Use in `emo`:** teacher model. `potion-multilingual-128M` was distilled from
  `bge-m3` (via Model2Vec / Tokenlearn), so `emo`'s semantic stream derives from it.

### Model2Vec / Tokenlearn — MinishLab
- **Source:** [github.com/MinishLab/model2vec](https://github.com/MinishLab/model2vec)
- **License:** MIT
- **Use in `emo`:** the static-embedding distillation method behind the semantic stream.

## Training data

### Unicode CLDR — Unicode, Inc.
- **Source:** [cldr.unicode.org](https://cldr.unicode.org) (emoji annotations / keywords).
- **License:** Unicode License Agreement — Data Files and Software (UNICODE-DFS-2016).
- **Use in `emo`:** multilingual emoji keywords were used as grounding examples when
  building the training data. CLDR data is not redistributed in this repository.