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 15, 02:15 UTC
[Bugfix] Restrict Machete to only run on Hopper (#20830)
Signed-off-by: mgoin <mgoin64@gmail.com>
Michael Goin committed
Jul 13, 2025 at 02:34 UTC
6e2c176e1ffd99382d03bc31a454834ca2532f7f
1 file changed
+3
vllm/model_executor/layers/quantization/kernels/mixed_precision/machete.py
+3
@@ -32,6 +32,9 @@ class MacheteLinearKernel(MPLinearKernel):
32
if not current_platform.is_cuda():
33
return False, "Machete only supported on CUDA"
34
35
+ if not current_platform.is_device_capability(90):
36
+ return False, "Machete requires compute capability of 90 (Hopper)"
37
+
38
if c.has_g_idx and\
39
c.partition_weight_shape[0] != c.full_weight_shape[0]:
40
return False, "Act reordering currently not supported by Machete, "\