paraphrase-multilingual-MiniLM-L12-v2 — Core ML, int8
Converted from sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2 @ e8f8c211… to Core ML with int8 per-channel weights by WristNotes. This is a modified version of the original work (Apache-2.0 §4(b)).
MeaningEncoder.mlmodelc: the compiled Core ML encoder. It outputs token embeddings; apply mask-weighted mean pooling and L2 normalisation to get a 384-dimension sentence embedding.tokenizer.json: the XLM-R SentencePiece (Unigram) tokenizer, unchanged.
Used on-device by WristNotes for Search by Meaning; nothing is sent anywhere.
Credits
- Sentence Transformers model: Apache-2.0 (see
LICENSE). - Base model: microsoft/Multilingual-MiniLM-L12-H384, MIT.
- Tokenizer vocabulary: XLM-R (Facebook AI, fairseq), MIT.