[XPU] support MLA model on Intel GPU (#37143)

Signed-off-by: Kunshang Ji <kunshang.ji@intel.com>
This commit is contained in:
Kunshang Ji
2026-03-25 17:43:42 +08:00
committed by GitHub
parent 189ddefbfd
commit 14771f7150
4 changed files with 15 additions and 12 deletions
@@ -165,6 +165,16 @@ class QuantFP8(CustomOp):
# Fallback to CUDA implementation
return self.forward_cuda(x, scale, scale_ub)
def forward_xpu(
self,
x: torch.Tensor,
scale: torch.Tensor | None = None,
scale_ub: torch.Tensor | None = None,
use_triton: bool = False,
) -> tuple[torch.Tensor, torch.Tensor]:
# XPU can use same code path as CUDA.
return self.forward_cuda(x, scale, scale_ub, use_triton)
def forward_native(
self,
x: torch.Tensor,