backend/app/services/cefr_predictor.py

import os
from pathlib import Path
import joblib

MODEL_PATH = Path(os.getenv("CEFR_MODEL_PATH", "/app/ml/cefr_model.pkl"))

HF_REPO_ID = os.getenv("CEFR_MODEL_REPO", "Zinebhm/cefr-model-learnlanguage")
HF_FILENAME = os.getenv("CEFR_MODEL_FILENAME", "cefr_model.pkl")

class CEFRPredictor:
    def __init__(self):
        if not MODEL_PATH.exists():
            MODEL_PATH.parent.mkdir(parents=True, exist_ok=True)
            # download from Hugging Face
            from huggingface_hub import hf_hub_download
            downloaded = hf_hub_download(
                repo_id=HF_REPO_ID,
                filename=HF_FILENAME,
                repo_type="model",
            )
            # copy to expected path
            os.replace(downloaded, MODEL_PATH)

        self.model = joblib.load(MODEL_PATH)

    def predict(self, text: str):
        return self.model.predict([text])[0]