EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

fix MultiTokenizerWrapper name (#10049)

Browse Source 
Signed-off-by: huanglong <huanglong@linux.alibaba.com>
This commit is contained in:
Huang Long
2025-09-05 13:39:46 +08:00
committed by GitHub
parent 8b3b995ac9
commit f98366604b
4 changed files with 11 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
python/sglang/srt/managers/io_struct.py
View File

@@ -1195,7 +1195,7 @@ class MultiTokenizerRegisterReq:
@dataclass @dataclass
class MultiTokenizerWarpper: class MultiTokenizerWrapper:
worker_id: int worker_id: int
obj: Optional[Any] = None obj: Optional[Any] = None

4
python/sglang/srt/managers/multi_tokenizer_mixin.py
View File

@@ -34,7 +34,7 @@ from sglang.srt.managers.io_struct import (
BatchStrOut, BatchStrOut,
BatchTokenIDOut, BatchTokenIDOut,
MultiTokenizerRegisterReq, MultiTokenizerRegisterReq,
MultiTokenizerWarpper, MultiTokenizerWrapper,
) )
from sglang.srt.managers.tokenizer_manager import TokenizerManager, _Communicator from sglang.srt.managers.tokenizer_manager import TokenizerManager, _Communicator
from sglang.srt.server_args import PortArgs, ServerArgs from sglang.srt.server_args import PortArgs, ServerArgs
@@ -441,7 +441,7 @@ class MultiTokenizerRouter(TokenizerManager, MultiTokenizerMixin):
async def _distribute_result_to_workers(self, recv_obj): async def _distribute_result_to_workers(self, recv_obj):
"""Distribute result to corresponding workers based on rid""" """Distribute result to corresponding workers based on rid"""
if isinstance(recv_obj, MultiTokenizerWarpper): if isinstance(recv_obj, MultiTokenizerWrapper):
worker_ids = [recv_obj.worker_id] worker_ids = [recv_obj.worker_id]
recv_obj = recv_obj.obj recv_obj = recv_obj.obj
else: else:

8
python/sglang/srt/managers/scheduler.py
View File

@@ -85,7 +85,7 @@ from sglang.srt.managers.io_struct import (
LoadLoRAAdapterReqInput, LoadLoRAAdapterReqInput,
LoadLoRAAdapterReqOutput, LoadLoRAAdapterReqOutput,
MultiTokenizerRegisterReq, MultiTokenizerRegisterReq,
MultiTokenizerWarpper, MultiTokenizerWrapper,
OpenSessionReqInput, OpenSessionReqInput,
OpenSessionReqOutput, OpenSessionReqOutput,
ProfileReq, ProfileReq,
@@ -1096,13 +1096,13 @@ class Scheduler(
self.send_to_tokenizer.send_pyobj(abort_req) self.send_to_tokenizer.send_pyobj(abort_req)
continue continue
# If it is a MultiTokenizerWarpper, unwrap it and handle the inner request. # If it is a MultiTokenizerWrapper, unwrap it and handle the inner request.
if isinstance(recv_req, MultiTokenizerWarpper): if isinstance(recv_req, MultiTokenizerWrapper):
worker_id = recv_req.worker_id worker_id = recv_req.worker_id
recv_req = recv_req.obj recv_req = recv_req.obj
output = self._request_dispatcher(recv_req) output = self._request_dispatcher(recv_req)
if output is not None: if output is not None:
output = MultiTokenizerWarpper(worker_id, output) output = MultiTokenizerWrapper(worker_id, output)
self.send_to_tokenizer.send_pyobj(output) self.send_to_tokenizer.send_pyobj(output)
continue continue

8
python/sglang/srt/managers/tokenizer_manager.py
View File

@@ -94,7 +94,7 @@ from sglang.srt.managers.io_struct import (
LoadLoRAAdapterReqInput, LoadLoRAAdapterReqInput,
LoadLoRAAdapterReqOutput, LoadLoRAAdapterReqOutput,
LoRAUpdateResult, LoRAUpdateResult,
MultiTokenizerWarpper, MultiTokenizerWrapper,
OpenSessionReqInput, OpenSessionReqInput,
OpenSessionReqOutput, OpenSessionReqOutput,
ProfileReq, ProfileReq,
@@ -1118,7 +1118,7 @@ class TokenizerManager:
self, obj: UpdateWeightFromDiskReqInput self, obj: UpdateWeightFromDiskReqInput
) -> Tuple[bool, str]: ) -> Tuple[bool, str]:
if self.server_args.tokenizer_worker_num > 1: if self.server_args.tokenizer_worker_num > 1:
obj = MultiTokenizerWarpper(self.worker_id, obj) obj = MultiTokenizerWrapper(self.worker_id, obj)
self.send_to_scheduler.send_pyobj(obj) self.send_to_scheduler.send_pyobj(obj)
self.model_update_result = asyncio.Future() self.model_update_result = asyncio.Future()
if self.server_args.dp_size == 1: if self.server_args.dp_size == 1:
@@ -1339,7 +1339,7 @@ class TokenizerManager:
return None return None
if self.server_args.tokenizer_worker_num > 1: if self.server_args.tokenizer_worker_num > 1:
obj = MultiTokenizerWarpper(self.worker_id, obj) obj = MultiTokenizerWrapper(self.worker_id, obj)
self.send_to_scheduler.send_pyobj(obj) self.send_to_scheduler.send_pyobj(obj)
self.session_futures[obj.session_id] = asyncio.Future() self.session_futures[obj.session_id] = asyncio.Future()
@@ -2165,7 +2165,7 @@ class _Communicator(Generic[T]):
if obj: if obj:
if _Communicator.enable_multi_tokenizer: if _Communicator.enable_multi_tokenizer:
obj = MultiTokenizerWarpper(worker_id=os.getpid(), obj=obj) obj = MultiTokenizerWrapper(worker_id=os.getpid(), obj=obj)
self._sender.send_pyobj(obj) self._sender.send_pyobj(obj)
self._result_event = asyncio.Event() self._result_event = asyncio.Event()