[AMD][V4.1] Fix wo_b input flatten for packed fp8 activations on gfx950 (#42796)
This commit is contained in:
@@ -2621,7 +2621,7 @@ class MQALayer(MqaAttentionBase):
|
||||
and not should_skip_mlp_all_reduce()
|
||||
)
|
||||
o, _ = self.wo_b(
|
||||
o if isinstance(o, Mxfp8SwizzledInput) else o.flatten(1),
|
||||
o.flatten(1) if isinstance(o, torch.Tensor) else o,
|
||||
skip_all_reduce=defer_all_reduce,
|
||||
)
|
||||
if defer_all_reduce:
|
||||
|
||||
Reference in New Issue
Block a user