change setting

app.py

@@ -3,29 +3,51 @@ from fastapi.middleware.cors import CORSMiddleware
 from pydantic import BaseModel
 import torch
 import joblib
-import numpy as np
 from transformers import AutoTokenizer
 from model_inference import load_model, predict_from_input
+import os
 
 app = FastAPI(title="Question Difficulty/Discrimination Predictor")
 
-# CORS for Next.js frontend
+# CORS for frontend usage (Next.js, Streamlit, etc.)
 app.add_middleware(
     CORSMiddleware,
-    allow_origins=["http://localhost:3000"],
+    allow_origins=["*"],
     allow_credentials=True,
     allow_methods=["*"],
     allow_headers=["*"],
 )
 
-# Load model, encoder, and scaler once at startup
-model, device = load_model("assets/best_checkpoint_regression.pt")
-encoder = joblib.load("assets/onehot_encoder.pkl")
-scaler = joblib.load("assets/scaler.pkl")
+# Global variables – will be loaded at startup
+model = None
+device = None
+encoder = None
+scaler = None
+tok_mcq = None
+tok_clin = None
 
-# Tokenizers
-tok_mcq = AutoTokenizer.from_pretrained("microsoft/BiomedNLP-PubMedBERT-base-uncased-abstract")
-tok_clin = AutoTokenizer.from_pretrained("emilyalsentzer/Bio_ClinicalBERT")
+@app.on_event("startup")
+def load_all_resources():
+    """
+    ✅ Load model + tokenizers + encoders only once at startup.
+    Avoids slow import times & prevents “Space in Error”.
+    """
+    global model, device, encoder, scaler, tok_mcq, tok_clin
+
+    print("🚀 Loading model and dependencies...")
+
+    # Load model from local or Hugging Face
+    model, device = load_model("assets/best_checkpoint_regression.pt")
+
+    # Load pretrained scaler + encoder
+    encoder = joblib.load("assets/onehot_encoder.pkl")
+    scaler = joblib.load("assets/scaler.pkl")
+
+    # Load tokenizers lazily
+    tok_mcq = AutoTokenizer.from_pretrained("microsoft/BiomedNLP-PubMedBERT-base-uncased-abstract")
+    tok_clin = AutoTokenizer.from_pretrained("emilyalsentzer/Bio_ClinicalBERT")
+
+    print("✅ All resources successfully loaded.")
 
 # Input schema
 class QuestionInput(BaseModel):
@@ -45,6 +67,12 @@ def health():
 
 @app.post("/predict")
 def predict(input_data: QuestionInput):
+    """
+    ✅ Main prediction endpoint.
+    """
+    if model is None:
+        return {"error": "Model not loaded. Try again in a few seconds."}
+
     pred = predict_from_input(
         input_data.dict(), model, device,
         tok_mcq, tok_clin, encoder, scaler

space.yaml → runtime.yaml

@@ -4,7 +4,7 @@ runtime:
 
 app:
   file: app.py
-  entrypoint: "uvicorn app:app --host 0.0.0.0 --port 7860"
+  entrypoint: "uvicorn app:app --host 0.0.0.0 --port $PORT"
 
 hardware:
   accelerator: "cpu"