@@ -1311,6 +1311,9 @@ class DummyMetrics:
13111311 def __init__ (self ):
13121312 self .requests_number = Mock (inc = Mock ())
13131313 self .num_requests_waiting = Mock (inc = Mock ())
1314+ self .prompt_tokens_total = Mock (inc = Mock ())
1315+ self .request_prompt_tokens = Mock (observe = Mock ())
1316+ self .request_params_max_tokens = Mock (observe = Mock ())
13141317
13151318 def inc_value (self , name , value = 1 , labelvalues = None ):
13161319 getattr (self , name ).inc (value )
@@ -2727,6 +2730,8 @@ def close(self):
27272730 with patch ("fastdeploy.engine.common_engine.Request" ) as MockRequest :
27282731 mock_request = Mock ()
27292732 mock_request .metrics .scheduler_recv_req_time = 0
2733+ mock_request .prompt_token_ids_len = 2
2734+ mock_request .sampling_params = Mock (max_tokens = 16 )
27302735 MockRequest .from_dict .return_value = mock_request
27312736
27322737 with (
@@ -2756,6 +2761,8 @@ def close(self):
27562761 with patch ("fastdeploy.engine.common_engine.Request" ) as MockRequest :
27572762 mock_request = Mock ()
27582763 mock_request .metrics .scheduler_recv_req_time = 0
2764+ mock_request .prompt_token_ids_len = 2
2765+ mock_request .sampling_params = Mock (max_tokens = 16 )
27592766 MockRequest .from_dict .return_value = mock_request
27602767
27612768 with (
@@ -3359,6 +3366,9 @@ class DummyMetrics:
33593366 def __init__ (self ):
33603367 self .requests_number = Mock (inc = Mock ())
33613368 self .num_requests_waiting = Mock (inc = Mock ())
3369+ self .prompt_tokens_total = Mock (inc = Mock ())
3370+ self .request_prompt_tokens = Mock (observe = Mock ())
3371+ self .request_params_max_tokens = Mock (observe = Mock ())
33623372
33633373 def inc_value (self , name , value = 1 , labelvalues = None ):
33643374 getattr (self , name ).inc (value )
0 commit comments