Fixed model loading

app.py

@@ -11,10 +11,6 @@ from model_inference import load_model, predict_from_input, download_from_hf
 # ✅ Redirect HuggingFace & NLTK cache
 os.environ["TRANSFORMERS_CACHE"] = "/tmp/hf_cache"
 os.environ["HF_HOME"] = "/tmp/hf_cache"
-nltk_data_dir = "/tmp/nltk_data"
-os.makedirs(nltk_data_dir, exist_ok=True)
-nltk.data.path.append(nltk_data_dir)
-os.environ["NLTK_DATA"] = nltk_data_dir
 
 app = FastAPI(title="Question Difficulty/Discrimination Predictor")
 

model_inference.py

@@ -1,10 +1,18 @@
+import os
+import nltk
+
+# ✅ Force NLTK to save/download in /tmp (writeable in Hugging Face Spaces)
+nltk_data_dir = "/tmp/nltk_data"
+os.makedirs(nltk_data_dir, exist_ok=True)
+os.environ["NLTK_DATA"] = nltk_data_dir
+nltk.data.path = [nltk_data_dir]
+
 import torch
 import numpy as np
 import textstat
 from utils.preprocess import compute_text_features
 from model_architecture import EnsembleBertBiLSTMRegressor 
 from huggingface_hub import hf_hub_download
-import os
 
 HF_REPO = "hathimazman/sqb-predict"
 MODEL_CACHE = "/tmp/models"         # ✅ Writable on Hugging Face Spaces