xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-04-12 01:27:02 +08:00
Code Issues Packages Projects Releases Wiki Activity

update

Browse Source 
Signed-off-by: Tyler Michael Smith <tysmith@redhat.com>
This commit is contained in:
Tyler Michael Smith 2025-06-20 19:57:07 +00:00
parent b4f17e12a4
commit 850dafea92
1 changed files with 5 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
vllm/distributed/device_communicators/all2all.py
View File

@ -138,9 +138,11 @@ class DeepEPAll2AllManagerBase(All2AllManagerBase):
super().__init__(cpu_group)
self.handle_cache = Cache()
# This is the DeepEP default. Stick to it till we can establish
# reasonable defaults based on profiling.
self.num_sms = 20
# Use all SMs for all2all communication
# This will need to be adjusted for dual-batch overlap
device = self.dp_group.device
props = torch.cuda.get_device_properties(device)
self.num_sms = props.multi_processor_count
def get_handle(self, kwargs):
raise NotImplementedError