src/retrieval/embedder.py

from sentence_transformers import SentenceTransformer
from functools import lru_cache
from src.config import Settings

@lru_cache(maxsize=1)
def _cached_model(name: str):
    # Load model without authentication token (for public models)
    return SentenceTransformer(name, token=False)

def get_embedder(settings: Settings):
    model = _cached_model(settings.embedding_model)
    def _encode(texts):
        return model.encode(texts, convert_to_numpy=True, normalize_embeddings=True)
    return _encode