apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-05 16:48:03 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

delete parallel_state.py (#3250)

Browse Source
This commit is contained in:
Yuanle Liu
2025-08-08 11:03:29 +08:00
committed by GitHub
parent fbdd6b0663
commit fac2f64837
3 changed files with 7 additions and 37 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
fastdeploy/distributed/communication.py
View File

@@ -20,8 +20,6 @@ import paddle
import paddle.distributed as dist import paddle.distributed as dist
from paddle.distributed import fleet from paddle.distributed import fleet
from fastdeploy.distributed.parallel_state import get_tensor_model_parallel_world_size
_TP_AR = None _TP_AR = None
@@ -39,7 +37,6 @@ def use_custom_allreduce(custom_all_reduce_max_bytes: int = 8192 * 1024):
hcg = fleet.get_hybrid_communicate_group() hcg = fleet.get_hybrid_communicate_group()
model_parallel_group = hcg.get_model_parallel_group() model_parallel_group = hcg.get_model_parallel_group()
global _TP_AR global _TP_AR
if get_tensor_model_parallel_world_size() > 1 and paddle.is_compiled_with_cuda():
from fastdeploy.distributed.custom_all_reduce import CustomAllreduce from fastdeploy.distributed.custom_all_reduce import CustomAllreduce
_TP_AR = CustomAllreduce(model_parallel_group, custom_all_reduce_max_bytes) _TP_AR = CustomAllreduce(model_parallel_group, custom_all_reduce_max_bytes)

31
fastdeploy/distributed/parallel_state.py
View File

@@ -1,31 +0,0 @@
"""
# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""
from paddle.distributed import fleet
def get_tensor_model_parallel_world_size():
"""Return world size for the tensor model parallel group."""
hcg = fleet.get_hybrid_communicate_group()
mp_size = hcg.get_model_parallel_world_size()
return mp_size
def get_tensor_model_parallel_rank():
"""Return my rank for the tensor model parallel group."""
hcg = fleet.get_hybrid_communicate_group()
mp_rank = hcg.get_model_parallel_rank()
return mp_rank

6
fastdeploy/worker/gpu_worker.py
View File

@@ -68,7 +68,11 @@ class GpuWorker(WorkerBase):
gc.collect() gc.collect()
paddle.device.cuda.empty_cache() paddle.device.cuda.empty_cache()
if self.parallel_config.enable_custom_all_reduce: if (
self.parallel_config.enable_custom_all_reduce
and self.parallel_config.tensor_parallel_size > 1
and paddle.is_compiled_with_cuda()
):
from fastdeploy.distributed.communication import use_custom_allreduce from fastdeploy.distributed.communication import use_custom_allreduce
use_custom_allreduce() use_custom_allreduce()