handling GPU warning and refactor

app.py

@@ -1,29 +1,31 @@
-import gradio as gr
+import sys
+import warnings
 import pandas as pd
+import torch
+import gradio as gr
+import spaces
 from contextretriever import WikipediaContextRetriever
 from processor import MCQInferenceProcessor
 
-import sys
-import warnings
 
-# Prevents Python from printing ignored garbage collection exceptions
 sys.unraisablehook = lambda unraisable: None
-
-# Hides background socket resource warnings
 warnings.filterwarnings("ignore", category=ResourceWarning)
 
 
-train_df = pd.read_csv("data/train.csv")
-test_df = pd.read_csv("data/test.csv")
-df_samples = pd.concat([train_df, test_df], ignore_index=True)
+try:
+    train_df = pd.read_csv("data/train.csv")
+    test_df = pd.read_csv("data/test.csv")
+    df_samples = pd.concat([train_df, test_df], ignore_index=True)
+except FileNotFoundError:
+    print("⚠️ Warning: Sample CSV data files not found in 'data/' directory. Random button will be empty.")
+    df_samples = pd.DataFrame()
 
 def get_random_question_from_local():
     """Pulls a random question row from the CSV and returns values for the UI inputs."""
     if df_samples.empty:
-        return "", "", "", "", "", ""
+        return "No samples available", "", "", "", "", ""
     
     random_row = df_samples.sample(n=1).iloc[0]
-    
     return (
         str(random_row.get("prompt", "")),
         str(random_row.get("A", "")),
@@ -33,22 +35,37 @@ def get_random_question_from_local():
         str(random_row.get("E", ""))
     )
 
-# Initialization outside the Gradio interface
 processor = MCQInferenceProcessor(model_repo_id="balamaniansp/bigru-map3loss-lr5e-5-layers3-drop02-vocab10k-seq512-emb256-hidden256-fold3")
 retriever = WikipediaContextRetriever(sample_size=1500)
 
+@spaces.GPU
 def run_mcq_pipeline(question, opt_a, opt_b, opt_c, opt_d, opt_e):
-    options_dict = {"A": opt_a, "B": opt_b, "C": opt_c, "D": opt_d, "E": opt_e}
+    # Fallback to avoid empty string matrix processing errors
+    options_dict = {
+        "A": opt_a if opt_a.strip() else "[Empty Option A]",
+        "B": opt_b if opt_b.strip() else "[Empty Option B]",
+        "C": opt_c if opt_c.strip() else "[Empty Option C]",
+        "D": opt_d if opt_d.strip() else "[Empty Option D]",
+        "E": opt_e if opt_e.strip() else "[Empty Option E]"
+    }
     
+    # Context extraction through RAG framework
     retrieved_blocks = retriever._get_context(question, top_n=1)
-    context_str = retrieved_blocks[0]["text"]
-    context_source = retrieved_blocks[0]["title"]
     
+    if retrieved_blocks and len(retrieved_blocks) > 0:
+        context_str = retrieved_blocks[0].get("text", "No context found.")
+        context_source = retrieved_blocks[0].get("title", "Unknown Source")
+    else:
+        context_str = "No applicable Wikipedia context retrieved."
+        context_source = "Fallback System"
+    
+    # Model inference call execution
     probabilities = processor.predict(context=context_str, question=question, options=options_dict)
     
     context_feedback = f"📚 Source: {context_source}\n\nFetched Context:\n{context_str}"
     return probabilities, context_feedback
 
+# Layout definition using standard Gradio Blocks
 with gr.Blocks(theme=gr.themes.Soft()) as demo:
     gr.Markdown("# 🧠 Academic Presentation: Open-Domain MCQ Solver (RAG)")
     gr.Markdown("Type a question manually, or click **🎲 Load Random Sample Question** to test an existing dataset question.")
@@ -70,6 +87,7 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
             output_labels = gr.Label(label="Model Probability Confidence Distribution")
             output_context = gr.Textbox(label="Retrieved Reference Framework Data", interactive=False, lines=10)
 
+    # Wire up action dependencies
     submit_btn.click(
         fn=run_mcq_pipeline,
         inputs=[input_q, input_a, input_b, input_c, input_d, input_e],