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

[torch.compile] Make inductor partition rules respect splitting_ops #25691 (#25845)

Browse Source 
Signed-off-by: baonudesifeizhai <baonudesifeizhai@gmail.com>
Signed-off-by: baonudesifeizhai <85092850+baonudesifeizhai@users.noreply.github.com>
Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
This commit is contained in:
baonudesifeizhai
2025-10-10 12:35:28 -04:00
committed by GitHub
parent e519281920
commit cddce79fda
9 changed files with 267 additions and 112 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
vllm/compilation/compiler_interface.py
View File

@@ -17,8 +17,6 @@ from vllm.compilation.counter import compilation_counter
from vllm.config import VllmConfig
from vllm.utils import is_torch_equal_or_newer
from .inductor_pass import pass_context
class CompilerInterface:
"""
@@ -209,13 +207,12 @@ class InductorStandaloneAdaptor(CompilerInterface):
from torch._inductor import standalone_compile
with pass_context(runtime_shape):
compiled_graph = standalone_compile(
graph,
example_inputs,
dynamic_shapes=dynamic_shapes,
options={"config_patches": current_config},
)
compiled_graph = standalone_compile(
graph,
example_inputs,
dynamic_shapes=dynamic_shapes,
options={"config_patches": current_config},
)
# Save the compiled artifact to disk in the specified path
assert key is not None
@@ -462,13 +459,12 @@ class InductorAdaptor(CompilerInterface):
torch._functorch.config.patch(enable_remote_autograd_cache=False)
)
with pass_context(runtime_shape):
compiled_graph = compile_fx(
graph,
example_inputs,
inner_compile=hijacked_compile_fx_inner,
config_patches=current_config,
)
compiled_graph = compile_fx(
graph,
example_inputs,
inner_compile=hijacked_compile_fx_inner,
config_patches=current_config,
)
# We treat VLLM_DISABLE_COMPILE_CACHE as the overall switch for torch
# compilation cache. So turn off the checks if we disable the