biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Bugfix] Fix for multinode crash on 4 PP (#6495)

Browse Source 
Signed-off-by: Muralidhar Andoorveedu <muralidhar.andoorveedu@centml.ai>
This commit is contained in:
Murali Andoorveedu
2024-07-17 04:25:10 -04:00
committed by GitHub
parent 5bf35a91e4
commit 5fa6e9876e
2 changed files with 17 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
tests/distributed/test_pipeline_parallel.py
View File

@@ -4,14 +4,12 @@ from ..utils import RemoteOpenAIServer
@pytest.mark.parametrize(
"TP_SIZE, PP_SIZE, EAGER_MODE, CHUNKED_PREFILL, MODEL_NAME",
[
"TP_SIZE, PP_SIZE, EAGER_MODE, CHUNKED_PREFILL, MODEL_NAME", [
(2, 2, 0, 1, "meta-llama/Meta-Llama-3-8B"),
(2, 2, 1, 0, "meta-llama/Meta-Llama-3-8B"),
(1, 3, 0, 0, "meta-llama/Meta-Llama-3-8B"),
# TODO: figure out why PP=4 tests are flaky
# (1, 4, 0, 1, "meta-llama/Meta-Llama-3-8B"),
# (1, 4, 1, 0, "meta-llama/Meta-Llama-3-8B"),
(1, 4, 0, 1, "meta-llama/Meta-Llama-3-8B"),
(1, 4, 1, 0, "meta-llama/Meta-Llama-3-8B"),
])
def test_compare_tp(TP_SIZE, PP_SIZE, EAGER_MODE, CHUNKED_PREFILL, MODEL_NAME):
pp_args = [