DeepSeek/DeepEP
1
0
Fork 0
mirror of https://github.com/deepseek-ai/DeepEP synced 2025-05-06 21:14:24 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Add NVSHMEM_IB_ENABLE_RELAXED_ORDERING

Browse Source
This commit is contained in:
Chenggang Zhao 2025-02-26 17:54:12 +08:00
parent 45f481b87b
commit 3885404ffb
1 changed files with 3 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
deep_ep/buffer.py
View File

@ -79,6 +79,9 @@ class Buffer:
# NOTES: NVSHMEM initialization requires at least 256 MiB # NOTES: NVSHMEM initialization requires at least 256 MiB
os.environ['NVSHMEM_CUMEM_GRANULARITY'] = f'{2 ** 29}' os.environ['NVSHMEM_CUMEM_GRANULARITY'] = f'{2 ** 29}'
# Disable PCIe relaxed ordering to avoid out-of-order messages
os.environ['NVSHMEM_IB_ENABLE_RELAXED_ORDERING'] = '0'
# NOTES: make sure AR (Adaptive Routing) is turned off while running normal kernels, as we cannot verify AR status in the code # NOTES: make sure AR (Adaptive Routing) is turned off while running normal kernels, as we cannot verify AR status in the code
# Synchronize using the root ID # Synchronize using the root ID
nvshmem_unique_ids = [None, ] * self.group_size nvshmem_unique_ids = [None, ] * self.group_size