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 14, 05:30 UTC
[compile] Disable dynamo guards check for AOT compilation. (#27288)
Signed-off-by: zhxchen17 <zhxchen17@fb.com> Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk>
Zhengxu Chen committed
Oct 28, 2025 at 08:58 UTC
a00d6254e998be472d8df9dc590784d6facf8d85
1 file changed
+1
vllm/compilation/decorators.py
+1
@@ -302,6 +302,7 @@ def _support_torch_compile(
302
start_monitoring_torch_compile(self.vllm_config)
303
loaded_fn = torch.compiler.load_compiled_function(f)
304
_verify_source_unchanged(loaded_fn.source_info(), self.vllm_config)
305
+ loaded_fn.disable_guard_check()
306
self.aot_compiled_fn = loaded_fn
307
except Exception as e:
308
if os.path.exists(aot_compilation_path):