[Frontend][torch.compile] CompilationConfig Overhaul (#20283): name change compilation level to compilation mode, deprecation compilation level (#26355)

Signed-off-by: morrison-turnansky <mturnans@redhat.com>
Signed-off-by: Morrison Turnansky <mturnans@redhat.com>
Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
This commit is contained in:
Morrison Turnansky
2025-10-15 02:51:16 +00:00
committed by GitHub
co-authored by Luka Govedič
parent e66d787bce
commit 96b9aa5aa0
42 changed files with 270 additions and 248 deletions
@@ -14,7 +14,7 @@ from vllm.compilation.counter import compilation_counter
from vllm.compilation.decorators import ignore_torch_compile, support_torch_compile
from vllm.config import (
CompilationConfig,
CompilationLevel,
CompilationMode,
CUDAGraphMode,
VllmConfig,
set_current_vllm_config,
@@ -199,10 +199,10 @@ def test_multi_graph_piecewise_compile(use_inductor_graph_partition: bool):
outputs = []
# piecewise compile
# vllmcompile compile
vllm_config = VllmConfig(
compilation_config=CompilationConfig(
level=CompilationLevel.PIECEWISE,
mode=CompilationMode.VLLM_COMPILE,
use_cudagraph=True,
splitting_ops=["silly::attention"],
cudagraph_capture_sizes=[1, 2],
@@ -251,7 +251,7 @@ def test_multi_graph_piecewise_compile(use_inductor_graph_partition: bool):
# no compile or cudagraph
vllm_config = VllmConfig(
compilation_config=CompilationConfig(
level=CompilationLevel.NO_COMPILATION,
mode=CompilationMode.NONE,
)
)
cudagraph_runtime_mode = CUDAGraphMode.NONE
@@ -280,7 +280,7 @@ def test_multi_graph_piecewise_compile(use_inductor_graph_partition: bool):
# piecewise compile without CUDA graph
vllm_config = VllmConfig(
compilation_config=CompilationConfig(
level=CompilationLevel.PIECEWISE,
mode=CompilationMode.VLLM_COMPILE,
use_cudagraph=False,
splitting_ops=["silly::attention"],
use_inductor_graph_partition=use_inductor_graph_partition,
+2 -2
View File
@@ -13,7 +13,7 @@ from vllm.compilation.counter import compilation_counter
from vllm.compilation.decorators import support_torch_compile
from vllm.config import (
CompilationConfig,
CompilationLevel,
CompilationMode,
CUDAGraphMode,
VllmConfig,
set_current_vllm_config,
@@ -61,7 +61,7 @@ def _run_simple_model(
):
vllm_config = VllmConfig(
compilation_config=CompilationConfig(
level=CompilationLevel.PIECEWISE,
mode=CompilationMode.VLLM_COMPILE,
use_cudagraph=True,
use_inductor=use_inductor,
splitting_ops=splitting_ops,
+5 -5
View File
@@ -21,7 +21,7 @@ from vllm.compilation.counter import compilation_counter
from vllm.compilation.decorators import support_torch_compile
from vllm.config import (
CompilationConfig,
CompilationLevel,
CompilationMode,
CUDAGraphMode,
VllmConfig,
set_current_vllm_config,
@@ -356,13 +356,13 @@ def test_toy_llama(
)
compile_config_no_compile = CompilationConfig(
level=CompilationLevel.NO_COMPILATION,
level=CompilationMode.NONE,
cudagraph_mode=CUDAGraphMode.NONE,
backend="eager",
)
compile_config_no_split = CompilationConfig(
level=CompilationLevel.PIECEWISE,
level=CompilationMode.VLLM_COMPILE,
use_inductor_graph_partition=use_inductor_graph_partition,
cudagraph_mode=CUDAGraphMode.PIECEWISE,
backend=backend,
@@ -458,14 +458,14 @@ def benchmark():
for piecewise in [False, True]:
if piecewise:
compilation_config = CompilationConfig(
level=CompilationLevel.PIECEWISE,
mode=CompilationMode.VLLM_COMPILE,
use_cudagraph=True,
splitting_ops=["silly::attention"],
cudagraph_capture_sizes=cudagraph_sizes,
)
else:
compilation_config = CompilationConfig(
level=CompilationLevel.PIECEWISE,
mode=CompilationMode.VLLM_COMPILE,
cudagraph_capture_sizes=cudagraph_sizes,
)