fixed os

app.py

@@ -1,35 +1,34 @@
 from fastapi import FastAPI
 from fastapi.middleware.cors import CORSMiddleware
 from pydantic import BaseModel
-from model_inference import load_model, predict_from_input, ensure_model_file
+import torch
 import joblib
+import os
 from transformers import AutoTokenizer
+from model_inference import load_model, predict_from_input
 
-app = FastAPI(title="SQB Predictor API")
+# ✅ FIX: Set Hugging Face cache to a writable directory
+os.environ["TRANSFORMERS_CACHE"] = "/tmp/hf_cache"
+os.environ["HF_HOME"] = "/tmp/hf_cache"
+
+app = FastAPI(title="Question Difficulty/Discrimination Predictor")
 
 app.add_middleware(
     CORSMiddleware,
     allow_origins=["*"],
+    allow_credentials=True,
     allow_methods=["*"],
     allow_headers=["*"],
 )
 
-@app.on_event("startup")
-def load_all_resources():
-    print("🚀 Downloading model and dependencies...")
-
-    model_path = ensure_model_file("best_checkpoint_regression.pt")
-    encoder_path = ensure_model_file("onehot_encoder.pkl")
-    scaler_path = ensure_model_file("scaler.pkl")
-
-    global model, device, encoder, scaler, tok_mcq, tok_clin
+# ✅ Load model on startup
+model, device = load_model()
 
-    model, device = load_model(model_path)
-    encoder = joblib.load(encoder_path)
-    scaler = joblib.load(scaler_path)
+encoder = joblib.load("assets/onehot_encoder.pkl")
+scaler = joblib.load("assets/scaler.pkl")
 
-    tok_mcq = AutoTokenizer.from_pretrained("microsoft/BiomedNLP-PubMedBERT-base-uncased-abstract")
-    tok_clin = AutoTokenizer.from_pretrained("emilyalsentzer/Bio_ClinicalBERT")
+tok_mcq = AutoTokenizer.from_pretrained("microsoft/BiomedNLP-PubMedBERT-base-uncased-abstract")
+tok_clin = AutoTokenizer.from_pretrained("emilyalsentzer/Bio_ClinicalBERT")
 
 class QuestionInput(BaseModel):
     StemText: str
@@ -43,11 +42,13 @@ class QuestionInput(BaseModel):
     BloomLevel: str
 
 @app.get("/health")
-def health_check():
+def health():
     return {"status": "ok"}
 
 @app.post("/predict")
 def predict(input_data: QuestionInput):
-    return predict_from_input(
-        input_data.dict(), model, device, tok_mcq, tok_clin, encoder, scaler
+    pred = predict_from_input(
+        input_data.dict(), model, device,
+        tok_mcq, tok_clin, encoder, scaler
     )
+    return pred

model_inference.py

@@ -4,6 +4,7 @@ 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"