Rename --stream-output to --incremental-streaming-output (#20614)

python/sglang/srt/entrypoints/context.py

@@ -199,13 +199,13 @@ class StreamingHarmonyContext(HarmonyContext):
                 completion_tokens is not None
                 and len(output_token_ids) == completion_tokens
             ):
-                # Case 1: When --stream-output is not set.
+                # Case 1: When --incremental-streaming-output is not set.
                 # The output_ids contains all tokens generated so far.
                 # We only need to process the new tokens.
                 new_token_ids = output_token_ids[self.num_processed_tokens :]
                 self.num_processed_tokens = len(output_token_ids)
             else:
-                # Case 2: When --stream-output is set.
+                # Case 2: When --incremental-streaming-output is set.
                 # The output_ids contains only the new tokens.
                 new_token_ids = output_token_ids
                 self.num_processed_tokens += len(output_token_ids)

python/sglang/srt/managers/tokenizer_manager.py

@@ -1574,7 +1574,7 @@ class TokenizerManager(TokenizerCommunicatorMixin, TokenizerManagerMultiItemMixi
                 state.text += recv_obj.output_strs[i]
                 # Not all request types have `stream` (e.g., EmbeddingReqInput). Default to non-streaming.
                 is_stream = getattr(state.obj, "stream", False)
-                if self.server_args.stream_output and is_stream:
+                if self.server_args.incremental_streaming_output and is_stream:
                     state.output_ids.extend(recv_obj.output_ids[i])
                     output_token_ids = state.output_ids[state.last_output_offset :]
                     state.last_output_offset = len(state.output_ids)
@@ -1590,7 +1590,7 @@ class TokenizerManager(TokenizerCommunicatorMixin, TokenizerManagerMultiItemMixi
 
             elif isinstance(recv_obj, BatchTokenIDOutput):
                 is_stream = getattr(state.obj, "stream", False)
-                if self.server_args.stream_output and is_stream:
+                if self.server_args.incremental_streaming_output and is_stream:
                     state.output_ids.extend(recv_obj.output_ids[i])
                     output_token_ids = state.output_ids[state.last_output_offset :]
                     state.last_output_offset = len(state.output_ids)

python/sglang/srt/server_args.py

@@ -367,7 +367,7 @@ class ServerArgs:
     pp_max_micro_batch_size: Optional[int] = None
     pp_async_batch_depth: int = 0
     stream_interval: int = 1
-    stream_output: bool = False
+    incremental_streaming_output: bool = False
     enable_streaming_session: bool = False
     random_seed: Optional[int] = None
     constrained_json_whitespace_pattern: Optional[str] = None
@@ -3791,10 +3791,17 @@ class ServerArgs:
             help="The interval (or buffer size) for streaming in terms of the token length. A smaller value makes streaming smoother, while a larger value makes the throughput higher",
         )
         parser.add_argument(
-            "--stream-output",
+            "--incremental-streaming-output",
             action="store_true",
             help="Whether to output as a sequence of disjoint segments.",
         )
+        parser.add_argument(
+            "--stream-output",
+            action=DeprecatedStoreTrueAction,
+            dest="incremental_streaming_output",
+            new_flag="--incremental-streaming-output",
+            help="[Deprecated] Use --incremental-streaming-output instead.",
+        )
         parser.add_argument(
             "--enable-streaming-session",
             action="store_true",
@@ -6284,6 +6291,32 @@ class DeprecatedAction(argparse.Action):
         )
 
 
+class DeprecatedStoreTrueAction(argparse.Action):
+    """Deprecated flag that still stores True and prints a warning."""
+
+    def __init__(
+        self,
+        option_strings,
+        dest,
+        new_flag=None,
+        nargs=0,
+        const=True,
+        default=False,
+        **kwargs,
+    ):
+        self.new_flag = new_flag
+        super().__init__(
+            option_strings, dest, nargs=nargs, const=const, default=default, **kwargs
+        )
+
+    def __call__(self, parser, namespace, values, option_string=None):
+        replacement = f" Use '{self.new_flag}' instead." if self.new_flag else ""
+        print_deprecated_warning(
+            f"'{option_string}' is deprecated and will be removed in a future release.{replacement}"
+        )
+        setattr(namespace, self.dest, True)
+
+
 def auto_choose_speculative_params(self: ServerArgs):
     """
     Automatically choose the parameters for speculative decoding.