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 10, 09:00 UTC
[CPU][BugFix] Disable AOT Compile for CPU (#32037)
Signed-off-by: Fadi Arafeh <fadi.arafeh@arm.com>
Fadi Arafeh committed
Jan 11, 2026 at 07:15 UTC
9103ed16967390f3bbd6df104dcd162db43d3148
1 file changed
+6
-1
vllm/envs.py
+6
-1
@@ -288,11 +288,16 @@ def use_aot_compile() -> bool:
288
from vllm.model_executor.layers.batch_invariant import (
289
vllm_is_batch_invariant,
290
)
291
+ from vllm.platforms import current_platform
292
from vllm.utils.torch_utils import is_torch_equal_or_newer
293
294
default_value = (
295
"1"
295
- if is_torch_equal_or_newer("2.10.0.dev") and not disable_compile_cache()
296
+ if is_torch_equal_or_newer("2.10.0.dev")
297
+ and not disable_compile_cache()
298
+ # Disabling AOT_COMPILE for CPU
299
+ # See: https://github.com/vllm-project/vllm/issues/32033
300
+ and not current_platform.is_cpu()
301
else "0"
302
)
303