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
[Compressed Tensors] Always clone output for compile robustness (#26849)
Signed-off-by: Kyle Sayers <kylesayrs@gmail.com> Co-authored-by: Michael Goin <mgoin64@gmail.com>
Kyle Sayers committed
Oct 16, 2025 at 15:29 UTC
a5464dcf92bba8dfd052fc79bfc40e08aee515d9
1 file changed
+1
-1
vllm/model_executor/layers/quantization/compressed_tensors/transform/linear.py
+1
-1
@@ -163,7 +163,7 @@ class CompressedTensorsLinearTransformMethod(LinearMethodBase):
163
if self.output_transform is not None:
164
for part_id, (start, length) in enumerate(self.partition_ranges):
165
x[:, start : start + length] = self.output_transform(
166
- x[:, start : start + length].contiguous(), part_id=part_id
166
+ x[:, start : start + length].clone(), part_id=part_id
167
)
168
169
return x