enable metrics by default (#4108)

* enable metrics by default

* fix according to reviewer comments

Author: lvhan028

lmdeploy/cli/serve.py

@@ -110,7 +110,7 @@ def add_parser_api_server():
         quant_policy = ArgumentHelper.quant_policy(pt_group)
         model_format = ArgumentHelper.model_format(pt_group)
         hf_overrides = ArgumentHelper.hf_overrides(pt_group)
-        enable_metrics = ArgumentHelper.enable_metrics(pt_group)
+        disable_metrics = ArgumentHelper.disable_metrics(pt_group)
         ArgumentHelper.dp(pt_group)
         ArgumentHelper.ep(pt_group)
         ArgumentHelper.enable_microbatch(pt_group)
@@ -135,7 +135,7 @@ def add_parser_api_server():
         tb_group._group_actions.append(quant_policy)
         tb_group._group_actions.append(model_format)
         tb_group._group_actions.append(hf_overrides)
-        tb_group._group_actions.append(enable_metrics)
+        tb_group._group_actions.append(disable_metrics)
         ArgumentHelper.rope_scaling_factor(tb_group)
         ArgumentHelper.num_tokens_per_iter(tb_group)
         ArgumentHelper.max_prefill_iters(tb_group)
@@ -217,7 +217,7 @@ def api_server(args):
                 max_prefill_token_num=args.max_prefill_token_num,
                 enable_microbatch=args.enable_microbatch,
                 enable_eplb=args.enable_eplb,
-                enable_metrics=args.enable_metrics,
+                enable_metrics=not args.disable_metrics,
                 role=EngineRole[args.role],
                 migration_backend=MigrationBackend[args.migration_backend],
                 model_format=args.model_format,
@@ -245,7 +245,7 @@ def api_server(args):
                                                    num_tokens_per_iter=args.num_tokens_per_iter,
                                                    max_prefill_iters=args.max_prefill_iters,
                                                    communicator=args.communicator,
-                                                   enable_metrics=args.enable_metrics,
+                                                   enable_metrics=not args.disable_metrics,
                                                    hf_overrides=args.hf_overrides)
         chat_template_config = get_chat_template(args.chat_template, args.model_path)
 

lmdeploy/cli/utils.py

@@ -594,9 +594,12 @@ def enable_eplb(parser):
         return parser.add_argument('--enable-eplb', action='store_true', help='enable eplb for specified model')
 
     @staticmethod
-    def enable_metrics(parser):
-        """Add argument enable_metrics to parser."""
-        return parser.add_argument('--enable-metrics', action='store_true', default=False, help='enable metrics system')
+    def disable_metrics(parser):
+        """Add argument disable_metrics to parser."""
+        return parser.add_argument('--disable-metrics',
+                                   action='store_true',
+                                   default=False,
+                                   help='disable metrics system')
 
     # For Disaggregation
     @staticmethod
@@ -623,7 +626,7 @@ def disable_vision_encoder(parser):
         return parser.add_argument('--disable-vision-encoder',
                                    action='store_true',
                                    default=False,
-                                   help='enable metrics system')
+                                   help='disable multimodal encoder')
 
     @staticmethod
     def logprobs_mode(parser):

lmdeploy/messages.py

@@ -259,7 +259,7 @@ class TurbomindEngineConfig:
     empty_init: bool = False
     communicator: str = 'nccl'
     hf_overrides: Optional[Dict[str, Any]] = None
-    enable_metrics: bool = False
+    enable_metrics: bool = True
 
     def __post_init__(self):
         """Check input validation."""
@@ -372,7 +372,7 @@ class PytorchEngineConfig:
     enable_mp_engine: bool = False
     mp_engine_backend: str = 'mp'
     model_format: str = None
-    enable_metrics: bool = False
+    enable_metrics: bool = True
     hf_overrides: Optional[Dict[str, Any]] = None
     disable_vision_encoder: bool = False
     logprobs_mode: str = None