Rename max_micro_batch_size -> pp_max_micro_batch_size (#11279)

docs/advanced_features/server_arguments.md

@@ -136,7 +136,7 @@ Please consult the documentation below and [server_args.py](https://github.com/s
 | `--device` | The device to use ('cuda', 'xpu', 'hpu', 'npu', 'cpu'). Defaults to auto-detection if not specified. | None |
 | `--tp-size` | The tensor parallelism size. | 1 |
 | `--pp-size` | The pipeline parallelism size. | 1 |
-| `--max-micro-batch-size` | The maximum micro batch size in pipeline parallelism. | None |
+| `--pp-max-micro-batch-size` | The maximum micro batch size in pipeline parallelism. | None |
 | `--stream-interval` | 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. | 1 |
 | `--stream-output` | Whether to output as a sequence of disjoint segments. | False |
 | `--random-seed` | The random seed. | None |

python/sglang/srt/entrypoints/http_server.py

@@ -494,7 +494,7 @@ async def get_load():
 
 
 # example usage:
-# curl -s -X POST http://localhost:30000/set_internal_state -H "Content-Type: application/json" -d '{"server_args": {"max_micro_batch_size": 8}}'
+# curl -s -X POST http://localhost:30000/set_internal_state -H "Content-Type: application/json" -d '{"server_args": {"pp_max_micro_batch_size": 8}}'
 @app.api_route("/set_internal_state", methods=["POST", "PUT"])
 async def set_internal_state(obj: SetInternalStateReq, request: Request):
     res = await _global_state.tokenizer_manager.set_internal_state(obj)

python/sglang/srt/managers/schedule_batch.py

@@ -97,7 +97,7 @@ GLOBAL_SERVER_ARGS_KEYS = [
     "ep_num_redundant_experts",
     "enable_nan_detection",
     "flashinfer_mla_disable_ragged",
-    "max_micro_batch_size",
+    "pp_max_micro_batch_size",
     "disable_shared_experts_fusion",
     "sampling_backend",
     "speculative_accept_threshold_single",

python/sglang/srt/managers/scheduler.py

@@ -464,8 +464,8 @@ class Scheduler(
             _,
             _,
         ) = self.tp_worker.get_worker_info()
-        if global_server_args_dict["max_micro_batch_size"] is None:
-            global_server_args_dict["max_micro_batch_size"] = max(
+        if global_server_args_dict["pp_max_micro_batch_size"] is None:
+            global_server_args_dict["pp_max_micro_batch_size"] = max(
                 self.max_running_requests // server_args.pp_size, 1
             )
 
@@ -1802,7 +1802,7 @@ class Scheduler(
         return ret
 
     def get_num_allocatable_reqs(self, running_bs):
-        res = global_server_args_dict["max_micro_batch_size"] - running_bs
+        res = global_server_args_dict["pp_max_micro_batch_size"] - running_bs
         if self.pp_size > 1:
             res = min(res, self.req_to_token_pool.available_size())
         return res
@@ -2510,7 +2510,7 @@ class Scheduler(
         server_args_dict = recv_req.server_args
         args_allow_update = set(
             [
-                "max_micro_batch_size",
+                "pp_max_micro_batch_size",
                 "speculative_accept_threshold_single",
                 "speculative_accept_threshold_acc",
             ]
@@ -2521,7 +2521,7 @@ class Scheduler(
                 logging.warning(f"Updating {k} is not supported.")
                 if_success = False
                 break
-            elif k == "max_micro_batch_size" and (
+            elif k == "pp_max_micro_batch_size" and (
                 v > self.max_running_requests // self.pp_size or v < 1
             ):
                 logging.warning(

python/sglang/srt/server_args.py

@@ -205,7 +205,7 @@ class ServerArgs:
     device: Optional[str] = None
     tp_size: int = 1
     pp_size: int = 1
-    max_micro_batch_size: Optional[int] = None
+    pp_max_micro_batch_size: Optional[int] = None
     stream_interval: int = 1
     stream_output: bool = False
     random_seed: Optional[int] = None
@@ -1599,9 +1599,9 @@ class ServerArgs:
             help="The pipeline parallelism size.",
         )
         parser.add_argument(
-            "--max-micro-batch-size",
+            "--pp-max-micro-batch-size",
             type=int,
-            default=ServerArgs.max_micro_batch_size,
+            default=ServerArgs.pp_max_micro_batch_size,
             help="The maximum micro batch size in pipeline parallelism.",
         )
         parser.add_argument(