modules/embed.py

from sentence_transformers import SentenceTransformer
import spaces

# Load embedding model
embedder = SentenceTransformer("sentence-transformers/all-mpnet-base-v2")
embedder.to("cuda")

@spaces.GPU(duration_s=10)  # Embedding typically takes very little time
def embed_text(text):
    embedding = embedder.encode(text, convert_to_tensor=True, device="cuda")
    return embedding.cpu().numpy()