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 11, 23:45 UTC
[bugfix] Pass globals to aot_compiled function (#29428)
Signed-off-by: angelayi <yiangela7@gmail.com>
Angela Yi committed
Dec 5, 2025 at 08:54 UTC
e7296b08da66b4c79eb43d0932a3d8628178d036
1 file changed
+3
-1
vllm/compilation/decorators.py
+3
-1
@@ -409,7 +409,9 @@ def _support_torch_compile(
409
open(aot_compilation_path, "rb") as f,
410
):
411
start_monitoring_torch_compile(self.vllm_config)
412
- loaded_fn = torch.compiler.load_compiled_function(f)
412
+ loaded_fn = torch.compiler.load_compiled_function(
413
+ f, f_globals=self.forward.__globals__
414
+ )
415
_verify_source_unchanged(loaded_fn.source_info(), self.vllm_config)
416
loaded_fn.disable_guard_check()
417
self.aot_compiled_fn = loaded_fn