biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Bugfix] Restrict Machete to only run on Hopper (#20830)

Browse Source 
Signed-off-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
Michael Goin
2025-07-13 02:34:40 +09:00
committed by GitHub
parent a86754a12b
commit 6e2c176e1f
1 changed files with 3 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
vllm/model_executor/layers/quantization/kernels/mixed_precision/machete.py
View File

@@ -32,6 +32,9 @@ class MacheteLinearKernel(MPLinearKernel):
if not current_platform.is_cuda(): if not current_platform.is_cuda():
return False, "Machete only supported on CUDA" return False, "Machete only supported on CUDA"
if not current_platform.is_device_capability(90):
return False, "Machete requires compute capability of 90 (Hopper)"
if c.has_g_idx and\ if c.has_g_idx and\
c.partition_weight_shape[0] != c.full_weight_shape[0]: c.partition_weight_shape[0] != c.full_weight_shape[0]:
return False, "Act reordering currently not supported by Machete, "\ return False, "Act reordering currently not supported by Machete, "\