Update app.py

app.py

@@ -13,10 +13,10 @@ 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))
+        return np.array(json.loads(x), dtype=np.float32)  # Ensure the array is of type float32
     except json.JSONDecodeError as e:
         print(f"Error decoding JSON: {e}")
-        return np.array([])  # 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)
@@ -25,7 +25,7 @@ data['embedding'] = data['embedding'].apply(safe_json_loads)
 data = data[data['embedding'].apply(lambda x: x.size > 0)]
 
 # Initialize FAISS index
-dimension = len(data['embedding'][0])
+dimension = len(data['embedding'].iloc[0])
 res = faiss.StandardGpuResources()  # use a single GPU
 
 # Create FAISS index
@@ -35,7 +35,9 @@ if faiss.get_num_gpus() > 0:
 else:
     gpu_index = faiss.IndexFlatL2(dimension)  # fall back to CPU
 
-gpu_index.add(np.stack(data['embedding'].values))
+# Ensure embeddings are stacked as float32
+embeddings = np.vstack(data['embedding'].values).astype(np.float32)
+gpu_index.add(embeddings)
 
 # Check if GPU is available
 device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
@@ -52,7 +54,7 @@ def embed_question(question, model, tokenizer):
     inputs = tokenizer(question, return_tensors='pt').to(device)
     with torch.no_grad():
         outputs = model(**inputs)
-    return outputs.last_hidden_state.mean(dim=1).cpu().numpy()
+    return outputs.last_hidden_state.mean(dim=1).cpu().numpy().astype(np.float32)
 
 # Function to retrieve the relevant document and generate a response
 @spaces.GPU(duration=120)