xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-14 17:05:36 +08:00
Code Issues Packages Projects Releases Wiki Activity

[ci][distributed] fix flaky tests (#6806)

Browse Source
This commit is contained in:
youkaichao 2024-07-25 17:44:09 -07:00 committed by GitHub
parent 1adddb14bf
commit 443c7cf4cf
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 8 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
tests/distributed/test_pipeline_parallel.py
View File

@ -1,3 +1,10 @@
"""
WARNING: This test runs in both single-node (4 GPUs) and multi-node
(2 node with 2 GPUs each) modes. If the test only uses 2 GPUs, it is
important to set the distributed backend to "mp" to avoid Ray scheduling
all workers in a node other than the head node, which can cause the test
to fail.
"""
import os import os
import pytest import pytest
@ -78,7 +85,7 @@ def test_pp_cudagraph(PP_SIZE, MODEL_NAME, ATTN_BACKEND):
"--pipeline-parallel-size", "--pipeline-parallel-size",
str(PP_SIZE), str(PP_SIZE),
"--distributed-executor-backend", "--distributed-executor-backend",
"ray", "mp",
] ]
os.environ["VLLM_ATTENTION_BACKEND"] = ATTN_BACKEND os.environ["VLLM_ATTENTION_BACKEND"] = ATTN_BACKEND