Update app.py

app.py

@@ -34,19 +34,28 @@ def generate_advice(extracted_data):
         for item in extracted_data:
             # Prepare the query string
             query = f"{item['Component']} {item['Status']}"
-            
-            # Generate query embedding and ensure it’s a 2D numpy array
+            print(f"Processing query: {query}")  # Debug print
+
+            # Generate query embedding and force it into 2D float32 numpy array
             query_embedding = embedding_model.encode([query])
-            query_embedding = np.array(query_embedding, dtype="float32").reshape(1, -1)
+            if not isinstance(query_embedding, np.ndarray):
+                query_embedding = np.array(query_embedding, dtype="float32")
+
+            query_embedding = query_embedding.reshape(1, -1).astype("float32")
 
-            # Check query embedding shape against FAISS index
+            # Debugging the embedding
+            print(f"Query Embedding Shape: {query_embedding.shape}, FAISS Index Dim: {index.d}")
+
+            # Validate shape compatibility
             if query_embedding.shape[1] != index.d:
                 raise ValueError(
                     f"Embedding dimension mismatch: Query ({query_embedding.shape[1]}), Index ({index.d})"
                 )
 
-            # Search for the closest match in the FAISS index
+            # Search for the closest match
             _, idx = index.search(query_embedding, 1)
+            print(f"FAISS Search Result: {idx}")  # Debug print
+
             best_match = kb[idx[0][0]]
 
             # Prepare the LLM prompt
@@ -73,7 +82,7 @@ def generate_advice(extracted_data):
                 return_tensors="pt",
             )
 
-            # Generate response using LLaMA
+            # Generate response
             output = llm.generate(
                 input_ids=input_text_with_your_role,
                 max_length=150,
@@ -81,7 +90,7 @@ def generate_advice(extracted_data):
             )
             advice = tokenizer.decode(output[0], skip_special_tokens=True).strip()
 
-            # Append results
+            # Append result
             recommendations.append({
                 "Component": item["Component"],
                 "Advice": advice
@@ -90,6 +99,7 @@ def generate_advice(extracted_data):
         return recommendations
 
     except Exception as e:
+        print(f"Error occurred: {str(e)}")  # Debugging error
         return [{"error": f"Exception occurred: {str(e)}"}]
 
 # Gradio app with LLM integration