import asyncio from functools import lru_cache from huggingface_hub import InferenceClient from app.core.config import get_settings class EmbeddingService: def __init__(self) -> None: settings = get_settings() if not settings.hf_api_token: raise ValueError("HF_API_TOKEN is required for Hugging Face Inference API.") self._client = InferenceClient(token=settings.hf_api_token) self._model = settings.hf_embedding_model async def embed(self, text: str) -> list[float]: loop = asyncio.get_event_loop() return await loop.run_in_executor( None, lambda: self._client.feature_extraction(text, model=self._model).tolist(), ) async def embed_batch(self, texts: list[str]) -> list[list[float]]: loop = asyncio.get_event_loop() return await loop.run_in_executor( None, lambda: self._client.feature_extraction(texts, model=self._model).tolist(), ) @lru_cache(maxsize=1) def get_embedding_service() -> EmbeddingService: return EmbeddingService()