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

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

Browse Source 
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
2025-10-14 22:51:16 -04:00
committed by GitHub
parent e66d787bce
commit 96b9aa5aa0
42 changed files with 270 additions and 248 deletions
Download Patch File Download Diff File Expand all files Collapse all files

39
tests/v1/cudagraph/test_cudagraph_mode.py
View File

@@ -10,7 +10,7 @@ import pytest
from tests.utils import wait_for_gpu_memory_to_clear
from tests.v1.attention.utils import full_cg_backend_configs as backend_configs
from vllm import LLM
from vllm.config import CompilationConfig
from vllm.config import CompilationConfig, CompilationMode
from vllm.platforms import current_platform
@@ -73,7 +73,7 @@ def test_backend_and_cudagraph_mode_combo(backend_name, cudagraph_mode, supporte
gpu_memory_utilization=0.45,
max_model_len=1024,
compilation_config=CompilationConfig(
level=3, cudagraph_mode=cudagraph_mode
mode=CompilationMode.VLLM_COMPILE, cudagraph_mode=cudagraph_mode
),
)
llm.generate(["Hello, my name is"] * 10)
@@ -90,32 +90,27 @@ def test_backend_and_cudagraph_mode_combo(backend_name, cudagraph_mode, supporte
)
# test cudagraph_mode with different compilation level.
# (backend_name, cudagraph_mode, compilation_level, supported)
# test cudagraph_mode with different compilation mode.
# (backend_name, cudagraph_mode, compilation_mode, supported)
combo_cases_2 = [
("FA2", "FULL", 0, True), # no compilation + full cudagraph
("FA2", "FULL", 3, True), # piecewise compilation + full cudagraph
("FA2", "PIECEWISE", 0, False), # no compilation + piecewise cudagraph
("FA2", "PIECEWISE", 3, True), # piecewise compilation + piecewise cudagraph
(
"FA2",
"FULL_AND_PIECEWISE",
0,
False,
), # piecewise cudagraph not supported without piecewise compilation
("FA2", "FULL_AND_PIECEWISE", 3, True),
("FA2", "FULL_DECODE_ONLY", 0, True),
("FA2", "FULL_DECODE_ONLY", 3, True),
("FA2", "NONE", 0, True), # no compilation + no cudagraph
("FA2", "NONE", 3, True), # piecewise compilation + no cudagraph
("FA2", "FULL", CompilationMode.NONE, True),
("FA2", "FULL", CompilationMode.VLLM_COMPILE, True),
("FA2", "PIECEWISE", CompilationMode.NONE, False),
("FA2", "PIECEWISE", CompilationMode.VLLM_COMPILE, True),
("FA2", "FULL_AND_PIECEWISE", CompilationMode.NONE, False),
("FA2", "FULL_AND_PIECEWISE", CompilationMode.VLLM_COMPILE, True),
("FA2", "FULL_DECODE_ONLY", CompilationMode.NONE, True),
("FA2", "FULL_DECODE_ONLY", CompilationMode.VLLM_COMPILE, True),
("FA2", "NONE", CompilationMode.NONE, True),
("FA2", "NONE", CompilationMode.VLLM_COMPILE, True),
]
@pytest.mark.parametrize(
"backend_name,cudagraph_mode,compilation_level,supported", combo_cases_2
"backend_name,cudagraph_mode,compilation_mode,supported", combo_cases_2
)
def test_cudagraph_compilation_combo(combo_case):
backend_name, cudagraph_mode, compilation_level, supported = combo_case
backend_name, cudagraph_mode, compilation_mode, supported = combo_case
env_vars = backend_configs[backend_name].env_vars
@@ -130,7 +125,7 @@ def test_cudagraph_compilation_combo(combo_case):
gpu_memory_utilization=0.45,
max_model_len=1024,
compilation_config=CompilationConfig(
level=compilation_level, cudagraph_mode=cudagraph_mode
mode=compilation_mode, cudagraph_mode=cudagraph_mode
),
)
llm.generate(["Hello, my name is"] * 10)