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

[FrontEnd] UNREVERT CompilationConfig overhaul (#20283): deprecate use_inductor in favor of backend, simplify custom_ops (#26502)

Browse Source 
Signed-off-by: morrison-turnansky <mturnans@redhat.com>
Signed-off-by: Morrison Turnansky <mturnans@redhat.com>
Signed-off-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
Co-authored-by: Jiangyun Zhu <riverclouds.zhu@qq.com>
This commit is contained in:
Morrison Turnansky
2025-10-13 18:47:16 -04:00
committed by GitHub
parent 7200a21cd1
commit e3fdb627d9
8 changed files with 153 additions and 86 deletions
Download Patch File Download Diff File Expand all files Collapse all files

54
tests/compile/test_basic_correctness.py
View File

@@ -77,14 +77,15 @@ class TestSetting:
method="encode",
),
# vision language model
TestSetting(
model="microsoft/Phi-3.5-vision-instruct",
model_args=["--trust-remote-code", "--max-model-len", "2048"],
pp_size=2,
tp_size=1,
attn_backend="FLASH_ATTN",
method="generate_with_image",
),
# See https://github.com/vllm-project/vllm/issues/26716.
# TestSetting(
# model="microsoft/Phi-3.5-vision-instruct",
# model_args=["--trust-remote-code", "--max-model-len", "2048"],
# pp_size=2,
# tp_size=1,
# attn_backend="FLASH_ATTN",
# method="generate_with_image",
# ),
],
)
def test_compile_correctness(
@@ -109,41 +110,46 @@ def test_compile_correctness(
with monkeypatch.context() as m:
m.setenv("VLLM_ATTENTION_BACKEND", attn_backend)
final_args = [
"--enforce-eager",
*model_args,
"-pp",
str(pp_size),
"-tp",
str(tp_size),
"-O.cudagraph_mode=none",
]
all_args: list[list[str]] = []
all_envs: list[dict[str, str] | None] = []
for level in [
CompilationLevel.NO_COMPILATION,
for comp_level in [
CompilationLevel.DYNAMO_AS_IS,
CompilationLevel.DYNAMO_ONCE,
CompilationLevel.PIECEWISE,
]:
all_args.append(final_args + [f"-O{level}"])
all_envs.append({})
for level in [CompilationLevel.NO_COMPILATION, comp_level]:
all_args.append(
final_args + [f"-O.level={level}", "-O.backend=inductor"]
)
# inductor will change the output, so we only compare if the output
# is close, not exactly the same.
compare_all_settings(
model,
all_args,
all_envs,
method=method if method != "generate" else "generate_close",
)
all_envs.clear()
all_args.clear()
# inductor will change the output, so we only compare if the output
# is close, not exactly the same.
compare_all_settings(
model,
all_args,
all_envs,
method=method if method != "generate" else "generate_close",
)
all_envs.clear()
all_args.clear()
for level in [
CompilationLevel.NO_COMPILATION,
CompilationLevel.DYNAMO_AS_IS,
CompilationLevel.DYNAMO_ONCE,
CompilationLevel.PIECEWISE,
]:
all_args.append(final_args + [f"-O{level}"])
all_args.append(final_args + [f"-O.level={level}", "-O.backend=eager"])
all_envs.append({})
all_envs.append({})
compare_all_settings(model, all_args * 3, all_envs, method=method)