apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-06 00:57:33 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[feature]2.2 custom_allreduce support cudagraph recapture (#4307)

Browse Source 
* custom_allreduce support cudagraph recapture

* delete code

* add shut_down/restart default group
This commit is contained in:
chen
2025-09-29 18:14:21 +08:00
committed by GitHub
parent 9cc2c99539
commit 63a03ee152
7 changed files with 31 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
fastdeploy/distributed/communication.py
View File

@@ -42,6 +42,12 @@ def use_custom_allreduce(custom_all_reduce_max_bytes: int = 8192 * 1024):
_TP_AR = CustomAllreduce(model_parallel_group, custom_all_reduce_max_bytes)
def custom_ar_clear_ipc_handles():
global _TP_AR
if _TP_AR is not None:
_TP_AR.clear_ipc_handles()
try:
@paddle.jit.marker.unified