add gc

app.py

@@ -5,6 +5,7 @@ from transformers import AutoTokenizer, AutoModel
 from duckduckgo_search import ddg
 import time
 import gradio as gr
+import gc
 
 def best_device():
     if torch.cuda.is_available():
@@ -64,6 +65,7 @@ def ask_question(question, local_content = '', web_content = ''):
     return response
 
 def on_click(question, kb_types):
+    gc.collect()
     if best_device() == 'cuda':
         torch.cuda.empty_cache()
         
@@ -86,6 +88,7 @@ def on_click(question, kb_types):
     result = ask_question(question, local_content, web_content)
     print(f'{result}\n\n----------------------------')
     
+    gc.collect()
     if best_device() == 'cuda':
         torch.cuda.empty_cache()
     return result