A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 4, 22:30 UTC
[Core] Pass donate_graph_module=True to standalone_compile (#39733)
Signed-off-by: Frederik Gossen <frgossen@meta.com>
Frederik Gossen committed
Apr 20, 2026 at 13:40 UTC
47fcb8ca68c1027ba7eb7a9056bb4596ee284221
1 file changed
+3
vllm/compilation/compiler_interface.py
+3
@@ -295,6 +295,9 @@ class InductorStandaloneAdaptor(CompilerInterface):
295
},
296
}
297
298
+ if is_torch_equal_or_newer("2.13.0.dev"):
299
+ compile_kwargs["donate_graph_module"] = True # type: ignore[assignment]
300
+
301
use_aot: bool = supports_aot and envs.VLLM_USE_MEGA_AOT_ARTIFACT
302
# only add 'aot' parameter if both supported and enabled...
303
# this will set bundled_autograd_cache