max model for A10G at fp16 seems to be 6.7B params

app.py

@@ -23,8 +23,8 @@ arg_dict = {
     # 'model_name_or_path': 'facebook/opt-125m', 
     # 'model_name_or_path': 'facebook/opt-1.3b', 
     # 'model_name_or_path': 'facebook/opt-2.7b', 
-    # 'model_name_or_path': 'facebook/opt-6.7b',
-    'model_name_or_path': 'facebook/opt-13b',
+    'model_name_or_path': 'facebook/opt-6.7b',
+    # 'model_name_or_path': 'facebook/opt-13b',
     'load_fp16' : True,
     # 'load_fp16' : False,
     'prompt_max_length': None,