Skip to content

[Bugfix][EP+DP] Fix internode check #19112

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Jun 4, 2025
Merged

[Bugfix][EP+DP] Fix internode check #19112

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
c528147
[Bugfix][EP+DP] Fix internode check
tlrmchlsmth Jun 3, 2025
38a7be4
rm hack from #19034
tlrmchlsmth Jun 3, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 0 additions & 6 deletions vllm/distributed/device_communicators/all2all.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -84,10 +84,6 @@ def __init__(self, cpu_group):
assert has_pplx, "pplx_kernels not found. Please follow https://github.com/vllm-project/vllm/blob/main/tools/ep_kernels/README.md to install pplx_kernels." # noqa
super().__init__(cpu_group)

# TODO(tms): Disable pplx-a2a intranode as it fails with the error:
# failed: cuda error /app/pplx/csrc/all_to_all/intranode.cpp:84 'invalid resource handle' # noqa
self.internode = True

if self.internode:
# inter-node communication needs nvshmem,
# intra-node communication uses p2p mapping directly
Expand Down Expand Up @@ -178,7 +174,6 @@ def _make_all2all_kwargs(self) -> dict[Any, Any]:
num_rdma_bytes = 1024 * 1024 * 1024
num_qps_per_rank = self.num_sms // 2
else:
assert self.intranode
num_rdma_bytes = 0
num_qps_per_rank = 1

Expand Down Expand Up @@ -243,7 +238,6 @@ def _make_all2all_kwargs(
if self.internode:
num_rdma_bytes = 1024 * 1024 * 1024
else:
assert self.intranode
num_rdma_bytes = deep_ep.Buffer.get_low_latency_rdma_size_hint(
num_max_dispatch_tokens_per_rank=max_num_tokens_per_dp_rank,
hidden=token_hidden_size,
Expand Down
3 changes: 1 addition & 2 deletions vllm/distributed/device_communicators/base_device_communicator.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,8 +49,7 @@ def __init__(self, cpu_group):

# all2all communication often has separate implementations for
# intra-node and inter-node communication
self.intranode = in_the_same_node_as(cpu_group, source_rank=0)
self.internode = not self.intranode
self.internode = not all(in_the_same_node_as(cpu_group, source_rank=0))

def get_handle(self, kwargs):
# get a handle for the all2all communication,
Expand Down