Update app.py

app.py

@@ -22,8 +22,11 @@ retriever = SentenceTransformer('all-MiniLM-L6-v2')
 
 # Load ONNX model for QA using optimum.onnxruntime
 # Model: Xenova/distilbert-base-uncased-distilled-squad (~260MB)
-# Use ORTModelForQuestionAnswering to load the ONNX model
-model = ORTModelForQuestionAnswering.from_pretrained("Xenova/distilbert-base-uncased-distilled-squad")
+# Specify file_name="model.onnx" to select the correct ONNX file
+model = ORTModelForQuestionAnswering.from_pretrained(
+    "Xenova/distilbert-base-uncased-distilled-squad",
+    file_name="model.onnx"
+)
 tokenizer = AutoTokenizer.from_pretrained("Xenova/distilbert-base-uncased-distilled-squad")
 qa_model = pipeline("question-answering", model=model, tokenizer=tokenizer, framework="ort")
 
@@ -89,7 +92,7 @@ def answer_question(question):
     
     # Compute cosine similarity with stored embeddings
     cos_scores = util.cos_sim(question_embedding, embeddings)[0]
-    top_k = min(2, len(corpus))  # Get top 3 or less if fewer paragraphs
+    top_k = min(2, len(corpus))  # Get top 2 or less if fewer paragraphs
     top_indices = np.argsort(-cos_scores)[:top_k]
     
     # Retrieve context (top 3 paragraphs)