[XPU] fix MoE triton backend in online fp8 quantization (#40109)

Signed-off-by: Yan Ma <yan.ma@intel.com>
This commit is contained in:
Yan Ma
2026-04-20 11:31:39 -04:00
committed by GitHub
parent 3a30eaa1d7
commit 595562651a
3 changed files with 13 additions and 4 deletions
@@ -1019,10 +1019,6 @@ class Fp8OnlineMoEMethod(Fp8MoEMethod):
layer.w2_weight[expert, :, :]
)
if current_platform.is_xpu():
w13.data = w13.transpose(-1, -2).contiguous()
w2.data = w2.transpose(-1, -2).contiguous()
# Shuffle weights to runtime format and setup kernel.
self._setup_kernel(
layer,