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, 03:45 UTC
[CI][Bugfix] Fix CI failures for "PyTorch Compilation Unit Tests" (#41940)
Signed-off-by: haosdent <haosdent@gmail.com>
haosdent committed
May 8, 2026 at 10:42 UTC
57c2f724c1215a2c27382248e017d74f3ec97bb1
1 file changed
+4
vllm/compilation/compiler_interface.py
+4
@@ -141,6 +141,10 @@ class AlwaysHitShapeEnv:
141
142
def __init__(self) -> None:
143
self.guards: list[Any] = []
144
+ # Read by torch._inductor.codecache.FxGraphHashDetails (torch>=2.11)
145
+ # to incorporate user-provided dynamic-shape hint overrides into the
146
+ # cache key. We never override hints, so an empty dict is correct.
147
+ self.var_to_hint_override: dict[Any, int] = {}
148
149
def evaluate_guards_expression(self, *args: Any, **kwargs: Any) -> Literal[True]:
150
return True