open-git/vllm-anthropic
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[torch.compile] fix simple inductor graph partition test (#27050)

Browse Source 
Signed-off-by: Boyuan Feng <boyuan@meta.com>
This commit is contained in:
Boyuan Feng
2025-10-16 18:09:36 -07:00
committed by GitHub
parent 4d4d6bad19
commit 17c540a993
1 changed files with 2 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
tests/compile/piecewise/test_simple.py
View File

@@ -142,8 +142,7 @@ def test_simple_piecewise_compile(use_inductor):
@torch.inference_mode() @torch.inference_mode()
@pytest.mark.parametrize("splitting_ops", [["silly::attention"], []]) def test_simple_inductor_graph_partition(monkeypatch):
def test_simple_inductor_graph_partition(splitting_ops, monkeypatch):
if not is_torch_equal_or_newer("2.9.0.dev"): if not is_torch_equal_or_newer("2.9.0.dev"):
pytest.skip("inductor graph partition is only available in PyTorch 2.9+") pytest.skip("inductor graph partition is only available in PyTorch 2.9+")
@@ -152,8 +151,7 @@ def test_simple_inductor_graph_partition(splitting_ops, monkeypatch):
monkeypatch.setenv("VLLM_DISABLE_COMPILE_CACHE", "1") monkeypatch.setenv("VLLM_DISABLE_COMPILE_CACHE", "1")
_run_simple_model( _run_simple_model(
# Inductor graph partition automatically resets splitting_ops to an empty list splitting_ops=["silly::attention"],
splitting_ops=splitting_ops,
use_inductor_graph_partition=True, use_inductor_graph_partition=True,
use_inductor=True, use_inductor=True,
# Since not splitting at fx graph level # Since not splitting at fx graph level