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 9, 13:33 UTC
[torch.compile] Stop assuming 32 bit indexing (#33113)
Signed-off-by: Richard Zou <zou3519@gmail.com>
Richard Zou committed
Jan 26, 2026 at 23:25 UTC
3b8f0fe59e67318c2c6bffb20eb54c6648082c04
1 file changed
+2
-1
vllm/config/compilation.py
+2
-1
@@ -280,9 +280,10 @@ class DynamicShapesConfig:
280
until this change picked up https://github.com/pytorch/pytorch/pull/169239.
281
"""
282
283
- assume_32_bit_indexing: bool = True
283
+ assume_32_bit_indexing: bool = False
284
"""
285
whether all tensor sizes can use 32 bit indexing.
286
+ `True` requires PyTorch 2.10+
287
"""
288
289
def compute_hash(self) -> str: