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 16, 08:15 UTC
[Bugfix] SymIntArrayRef expected to contain concrete integers (#10170)
Signed-off-by: Bill Nell <bill@neuralmagic.com>
bnellnm committed
Nov 8, 2024 at 17:44 UTC
10b67d865d92e376956345becafc249d4c3c0ab7
1 file changed
+2
-1
vllm/compilation/backends.py
+2
-1
@@ -329,7 +329,8 @@ class PiecewiseCompileInterpreter(torch.fx.Interpreter):
329
self.fake_mode.from_tensor(t) if isinstance(t, torch.Tensor) else t
330
for t in args
331
]
332
- return super().run(*fake_args)
332
+ with self.fake_mode:
333
+ return super().run(*fake_args)
334
335
def call_module(self, target: torch.fx.node.Target,
336
args: Tuple[torch.fx.node.Argument,