A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[torch.compile] fix simple inductor graph partition test (#27050)
Signed-off-by: Boyuan Feng <boyuan@meta.com>
Boyuan Feng committed
Oct 16, 2025 at 18:09 UTC
17c540a993af88204ad1b78345c8a865cf58ce44
1 file changed
+2
-4
tests/compile/piecewise/test_simple.py
+2
-4
@@ -142,8 +142,7 @@ def test_simple_piecewise_compile(use_inductor):
142
143
144
@torch.inference_mode()
145
-@pytest.mark.parametrize("splitting_ops", [["silly::attention"], []])
146
-def test_simple_inductor_graph_partition(splitting_ops, monkeypatch):
145
+def test_simple_inductor_graph_partition(monkeypatch):
146
if not is_torch_equal_or_newer("2.9.0.dev"):
147
pytest.skip("inductor graph partition is only available in PyTorch 2.9+")
148
@@ -152,8 +151,7 @@ def test_simple_inductor_graph_partition(splitting_ops, monkeypatch):
151
monkeypatch.setenv("VLLM_DISABLE_COMPILE_CACHE", "1")
152
153
_run_simple_model(
155
- # Inductor graph partition automatically resets splitting_ops to an empty list
156
- splitting_ops=splitting_ops,
154
+ splitting_ops=["silly::attention"],
155
use_inductor_graph_partition=True,
156
use_inductor=True,
157
# Since not splitting at fx graph level