Expand description
Model2Vec backend — static, multilingual embeddings via model2vec-rs.
Loads minishlab/potion-* checkpoints from the HuggingFace Hub (or a
local path), encodes batches of text through the precomputed token-vector
table, and returns L2-normalised vectors. All computation is CPU-only and
orders of magnitude faster than the BERT path — typically ms/text on a
laptop CPU with no GPU.
Structs§
- Model2
VecBackend - Static-embedding backend powered by
model2vec-rs.