diff --git a/python/sglang/srt/entrypoints/http_server.py b/python/sglang/srt/entrypoints/http_server.py
index cace646d1..87197e5b7 100644
--- a/python/sglang/srt/entrypoints/http_server.py
+++ b/python/sglang/srt/entrypoints/http_server.py
@@ -689,6 +689,7 @@ async def start_profile_async(obj: Optional[ProfileReqInput] = None):
         record_shapes=obj.record_shapes,
         profile_by_stage=obj.profile_by_stage,
         merge_profiles=obj.merge_profiles,
+        profile_prefix=obj.profile_prefix,
     )
     return Response(
         content="Start profiling.\n",
diff --git a/python/sglang/srt/managers/io_struct.py b/python/sglang/srt/managers/io_struct.py
index cdecebca3..01ffab062 100644
--- a/python/sglang/srt/managers/io_struct.py
+++ b/python/sglang/srt/managers/io_struct.py
@@ -1332,6 +1332,7 @@ class ProfileReqInput(BaseReq):
     record_shapes: Optional[bool] = None
     # Merge profiles from all ranks into a single trace
     merge_profiles: bool = False
+    profile_prefix: Optional[str] = None
 
 
 class ProfileReqType(Enum):
@@ -1352,6 +1353,7 @@ class ProfileReq(BaseReq):
     profile_id: Optional[str] = None
     # Merge profiles from all ranks into a single trace
     merge_profiles: bool = False
+    profile_prefix: Optional[str] = None
 
 
 @dataclass
diff --git a/python/sglang/srt/managers/scheduler_profiler_mixin.py b/python/sglang/srt/managers/scheduler_profiler_mixin.py
index 969158bd9..b230e90f7 100644
--- a/python/sglang/srt/managers/scheduler_profiler_mixin.py
+++ b/python/sglang/srt/managers/scheduler_profiler_mixin.py
@@ -54,6 +54,7 @@ class SchedulerProfilerMixin:
         profile_by_stage: bool,
         profile_id: str,
         merge_profiles: bool = False,
+        profile_prefix: str = "",
     ) -> ProfileReqOutput:
         if self.profile_in_progress:
             return ProfileReqOutput(
@@ -74,6 +75,7 @@ class SchedulerProfilerMixin:
         self.torch_profiler_record_shapes = record_shapes
         self.profiler_activities = activities
         self.profile_id = profile_id
+        self.profile_prefix = profile_prefix
 
         if start_step:
             self.profiler_start_forward_ct = max(start_step, self.forward_ct + 1)
@@ -327,6 +329,7 @@ class SchedulerProfilerMixin:
                     recv_req.profile_by_stage,
                     recv_req.profile_id,
                     recv_req.merge_profiles,
+                    recv_req.profile_prefix,
                 )
             else:
                 self.init_profile(
@@ -339,6 +342,7 @@ class SchedulerProfilerMixin:
                     recv_req.profile_by_stage,
                     recv_req.profile_id,
                     recv_req.merge_profiles,
+                    recv_req.profile_prefix,
                 )
                 return self.start_profile()
         else:
diff --git a/python/sglang/srt/managers/tokenizer_communicator_mixin.py b/python/sglang/srt/managers/tokenizer_communicator_mixin.py
index 107a7f6ca..70129ea8c 100644
--- a/python/sglang/srt/managers/tokenizer_communicator_mixin.py
+++ b/python/sglang/srt/managers/tokenizer_communicator_mixin.py
@@ -315,6 +315,7 @@ class TokenizerCommunicatorMixin:
         record_shapes: Optional[bool] = None,
         profile_by_stage: bool = False,
         merge_profiles: bool = False,
+        profile_prefix: Optional[str] = None,
     ):
         self.auto_create_handle_loop()
         env_with_stack: bool = get_bool_env_var("SGLANG_PROFILE_WITH_STACK", "true")
@@ -334,6 +335,7 @@ class TokenizerCommunicatorMixin:
             profile_by_stage=profile_by_stage,
             profile_id=str(time.time()),
             merge_profiles=merge_profiles,
+            profile_prefix=profile_prefix,
         )
         return await self._execute_profile(req)
 
diff --git a/python/sglang/srt/model_executor/forward_batch_info.py b/python/sglang/srt/model_executor/forward_batch_info.py
index a4f2e7025..53969bbf9 100644
--- a/python/sglang/srt/model_executor/forward_batch_info.py
+++ b/python/sglang/srt/model_executor/forward_batch_info.py
@@ -90,6 +90,27 @@ class ForwardMode(IntEnum):
     # Split Prefill for PD multiplexing
     SPLIT_PREFILL = auto()
 
+    def __str__(self):
+        if self == ForwardMode.EXTEND:
+            return "EXTEND"
+        elif self == ForwardMode.DECODE:
+            return "DECODE"
+        elif self == ForwardMode.MIXED:
+            return "MIXED"
+        elif self == ForwardMode.IDLE:
+            return "IDLE"
+        elif self == ForwardMode.TARGET_VERIFY:
+            return "TARGET_VERIFY"
+        elif self == ForwardMode.DRAFT_EXTEND:
+            return "DRAFT_EXTEND"
+        elif self == ForwardMode.DRAFT_EXTEND_V2:
+            return "DRAFT_EXTEND_V2"
+        elif self == ForwardMode.PREBUILT:
+            return "PREBUILT"
+        elif self == ForwardMode.SPLIT_PREFILL:
+            return "SPLIT_PREFILL"
+        return "UNKNOWN"
+
     def is_prefill(self):
         return self.is_extend()
 
diff --git a/python/sglang/test/send_one.py b/python/sglang/test/send_one.py
index cff8ec733..1a9ec1dac 100644
--- a/python/sglang/test/send_one.py
+++ b/python/sglang/test/send_one.py
@@ -37,6 +37,7 @@ class BenchArgs:
     profile: bool = False
     profile_steps: int = 3
     profile_by_stage: bool = False
+    profile_name_prefix: str = ""
 
     @staticmethod
     def add_cli_args(parser: argparse.ArgumentParser):
@@ -44,6 +45,9 @@ class BenchArgs:
         parser.add_argument("--port", type=int, default=BenchArgs.port)
         parser.add_argument("--batch-size", type=int, default=BenchArgs.batch_size)
         parser.add_argument("--temperature", type=float, default=BenchArgs.temperature)
+        parser.add_argument(
+            "--profile-name-prefix", type=str, default=BenchArgs.profile_name_prefix
+        )
         parser.add_argument(
             "--max-new-tokens", type=int, default=BenchArgs.max_new_tokens
         )
@@ -133,6 +137,7 @@ def send_one_prompt(args):
             None,
             None,
             args.profile_by_stage,
+            args.profile_name_prefix,
         )
 
     response = requests.post(