[Bugfix] compressed-tensors: allow int8 grouped WNA16 MoE on Marlin (#47154)
Signed-off-by: Joe Rowell <joerowell4@gmail.com> Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>
Joe Rowell committed
Jun 30, 2026 at 21:50 UTC
c8d2f3cb1485fcca725653fb92a445b6cc10ade7
1 file changed
-1
vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors_moe/compressed_tensors_moe_wna16_marlin.py
-1
@@ -80,7 +80,6 @@ class CompressedTensorsWNA16MarlinMoEMethod(CompressedTensorsMoEMethod):
80
else:
81
scale = kInt4StaticGroupScale
82
elif self.num_bits == 8:
83
- assert self.group_size == -1
83
scale = kInt8StaticGroupScale
84
else:
85
raise ValueError(