EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Add --enable-p2p-check option (#599)

Browse Source
This commit is contained in:
Liangsheng Yin
2024-07-06 23:34:10 -07:00
committed by GitHub
parent 26908d9568
commit 5304b4ef58
4 changed files with 13 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
python/sglang/srt/utils.py
View File

@@ -458,13 +458,8 @@ def monkey_patch_vllm_p2p_access_check(gpu_id: int):
NOTE: We assume the p2p access is always allowed, which can be wrong for some setups.
"""
# TODO: need a better check than just dev str name match
# compat: skip RTX 40 series as they do not have P2P feature and even checking for them may cause errors
device_name = torch.cuda.get_device_name(gpu_id)
if "RTX 40" not in device_name:
import vllm.distributed.device_communicators.custom_all_reduce_utils as tgt
setattr(tgt, "gpu_p2p_access_check", lambda *arg, **kwargs: True)
import vllm.distributed.device_communicators.custom_all_reduce_utils as tgt
setattr(tgt, "gpu_p2p_access_check", lambda *arg, **kwargs: True)
def monkey_patch_vllm_dummy_weight_loader():