xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-06-05 15:02:22 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix test_simple_inductor_graph_partition (#26522)

Browse Source 
Signed-off-by: Boyuan Feng <boyuan@meta.com>
This commit is contained in:
Boyuan Feng 2025-10-09 23:39:19 -07:00 committed by GitHub
parent 29255cfc3b
commit b545a0b207
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 13 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

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

@ -143,10 +143,14 @@ def test_simple_piecewise_compile(use_inductor):
@torch.inference_mode() @torch.inference_mode()
@pytest.mark.parametrize("splitting_ops", [["silly.attention"], []]) @pytest.mark.parametrize("splitting_ops", [["silly.attention"], []])
def test_simple_inductor_graph_partition(splitting_ops): 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+")
# disable compile cache so that we run separately for different splitting_ops
# and get the expected number of cudagraphs captured.
monkeypatch.setenv("VLLM_DISABLE_COMPILE_CACHE", "1")
_run_simple_model( _run_simple_model(
# Inductor graph partition automatically resets splitting_ops to an empty list # Inductor graph partition automatically resets splitting_ops to an empty list
splitting_ops=splitting_ops, splitting_ops=splitting_ops,

10
vllm/compilation/compiler_interface.py
View File

@ -332,7 +332,10 @@ class InductorAdaptor(CompilerInterface):
nonlocal file_path nonlocal file_path
compiled_fn = inductor_compiled_graph.current_callable compiled_fn = inductor_compiled_graph.current_callable
file_path = compiled_fn.__code__.co_filename # noqa file_path = compiled_fn.__code__.co_filename # noqa
if not file_path.startswith(self.base_cache_dir): if (
not file_path.startswith(self.base_cache_dir)
and compiled_fn.__closure__ is not None
):
# hooked in the align_inputs_from_check_idxs function # hooked in the align_inputs_from_check_idxs function
# in torch/_inductor/utils.py # in torch/_inductor/utils.py
for cell in compiled_fn.__closure__: for cell in compiled_fn.__closure__:
@ -359,7 +362,10 @@ class InductorAdaptor(CompilerInterface):
nonlocal file_path nonlocal file_path
compiled_fn = inductor_compiled_graph.current_callable compiled_fn = inductor_compiled_graph.current_callable
file_path = compiled_fn.__code__.co_filename # noqa file_path = compiled_fn.__code__.co_filename # noqa
if not file_path.startswith(self.base_cache_dir): if (
not file_path.startswith(self.base_cache_dir)
and compiled_fn.__closure__ is not None
):
# hooked in the align_inputs_from_check_idxs function # hooked in the align_inputs_from_check_idxs function
# in torch/_inductor/utils.py # in torch/_inductor/utils.py
for cell in compiled_fn.__closure__: for cell in compiled_fn.__closure__: