xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-11 10:55:30 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix] Fix CustomAllreduce nvlink topology detection (#3974)

Browse Source 
[Bugfix] Fix CustomAllreduce pcie nvlink topology detection (#3974) (#4159)
This commit is contained in:
Adam Tilghman 2024-04-18 15:32:47 -07:00 committed by GitHub
parent cd2f63fb36
commit 8f9c28fd40
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 4 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
vllm/distributed/device_communicators/custom_all_reduce.py
View File

@ -145,8 +145,10 @@ def _is_full_nvlink(rank, world_size):
for i in range(world_size): for i in range(world_size):
if i != rank: if i != rank:
try: try:
link_state = pynvml.nvmlDeviceGetNvLinkState(handle, i) peer_handle = pynvml.nvmlDeviceGetHandleByIndex(i)
if not link_state: p2p_status = pynvml.nvmlDeviceGetP2PStatus(
handle, peer_handle, pynvml.NVML_P2P_CAPS_INDEX_NVLINK)
if p2p_status != pynvml.NVML_P2P_STATUS_OK:
return False return False
except pynvml.NVMLError as error: except pynvml.NVMLError as error:
logger.info( logger.info(