Update app.py

app.py

@@ -17,10 +17,10 @@ data = pd.read_csv('RBDx10kstats.csv')
 # Function to safely convert JSON strings to numpy arrays
 def safe_json_loads(x):
     try:
-        return np.array(json.loads(x), dtype=np.float16)  # Ensure the array is of type float32
+        return np.array(json.loads(x), dtype=np.float32)  # Ensure the array is of type float32
     except json.JSONDecodeError as e:
         logging.error(f"Error decoding JSON: {e}")
-        return np.array([], dtype=np.float16)  # Return an empty array or handle it as appropriate
+        return np.array([], dtype=np.float32)  # Return an empty array or handle it as appropriate
 
 # Apply the safe_json_loads function to the embedding column
 data['embedding'] = data['embedding'].apply(safe_json_loads)
@@ -40,7 +40,7 @@ else:
     gpu_index = faiss.IndexFlatL2(dimension)  # fall back to CPU
 
 # Ensure embeddings are stacked as float32
-embeddings = np.vstack(data['embedding'].values).astype(np.float16)
+embeddings = np.vstack(data['embedding'].values).astype(np.float32)
 logging.debug(f"Embeddings shape: {embeddings.shape}, dtype: {embeddings.dtype}")
 gpu_index.add(embeddings)
 
@@ -61,7 +61,7 @@ def embed_question(question, model, tokenizer):
         logging.debug(f"Tokenized inputs: {inputs}")
         with torch.no_grad():
             outputs = model(**inputs)
-        embedding = outputs.last_hidden_state.mean(dim=1).cpu().numpy().astype(np.float16)
+        embedding = outputs.last_hidden_state.mean(dim=1).cpu().numpy().astype(np.float32)
         logging.debug(f"Question embedding shape: {embedding.shape}")
         logging.debug(f"Question embedding content: {embedding}")
         return embedding
@@ -78,16 +78,16 @@ def retrieve_and_generate(question):
         question_embedding = embed_question(question, model, tokenizer)
 
         # Ensure the embedding is in the correct format for FAISS search
-        question_embedding = question_embedding.astype(np.float16)
+        question_embedding = question_embedding.astype(np.float32)
 
         # Search in FAISS index
         try:
             logging.debug(f"Searching FAISS index with question embedding: {question_embedding}")
-            _, indices = gpu_index.search(question_embedding, k=1)
-            if indices.size == 0:
+            distances, indices = gpu_index.search(question_embedding, k=1)
+            if len(indices) == 0:
                 logging.error("No results found in FAISS search.")
                 return "No relevant document found."
-            logging.debug(f"Indices found: {indices}")
+            logging.debug(f"Indices found: {indices}, Distances: {distances}")
         except Exception as e:
             logging.error(f"Error during FAISS search: {e}")
             return f"An error occurred during search: {e}"