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

6
vllm/compilation/backends.py
View File

@@ -41,7 +41,7 @@ logger = init_logger(__name__)
def make_compiler(compilation_config: CompilationConfig) -> CompilerInterface:
if compilation_config.use_inductor:
if compilation_config.backend == "inductor":
# Use standalone compile only if requested, version is new enough,
# and the symbol actually exists in this PyTorch build.
if (
@@ -55,6 +55,10 @@ def make_compiler(compilation_config: CompilationConfig) -> CompilerInterface:
logger.debug("Using InductorAdaptor")
return InductorAdaptor()
else:
assert compilation_config.backend == "eager", (
"Custom backends not supported with CompilationLevel.PIECEWISE"
)
logger.debug("Using EagerAdaptor")
return EagerAdaptor()