Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
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
5 changes: 4 additions & 1 deletion tests/kernels/moe/deepep_utils.py
Original file line number Diff line number Diff line change
Expand Up @@ -4,6 +4,7 @@
"""
import dataclasses
import importlib
import os
import traceback
from typing import Callable, Optional

Expand All @@ -13,6 +14,8 @@
spawn) # pyright: ignore[reportPrivateImportUsage]
from typing_extensions import Concatenate, ParamSpec

from vllm.model_executor.layers.fused_moe.utils import find_free_port

has_deep_ep = importlib.util.find_spec("deep_ep") is not None
if has_deep_ep:
from vllm.model_executor.layers.fused_moe.deepep_ht_prepare_finalize import ( # noqa: E501
Expand Down Expand Up @@ -92,7 +95,7 @@ def parallel_launch(
world_size,
world_size,
0,
"tcp://localhost:29500",
f"tcp://{os.getenv('LOCALHOST', 'localhost')}:{find_free_port()}",
worker,
) + args,
nprocs=world_size,
Expand Down
9 changes: 9 additions & 0 deletions vllm/model_executor/layers/fused_moe/utils.py
Original file line number Diff line number Diff line change
@@ -1,5 +1,7 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import socket
from contextlib import closing
from math import prod
from typing import Optional

Expand Down Expand Up @@ -96,3 +98,10 @@ def _fp8_perm(m: torch.Tensor, idx: torch.Tensor) -> torch.Tensor:
return m.view(dtype=torch.uint8)[idx, ...].view(dtype=m.dtype)
else:
return m[idx, ...]


def find_free_port():
with closing(socket.socket(socket.AF_INET, socket.SOCK_STREAM)) as s:
s.bind(('', 0))
s.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
return s.getsockname()[1]