diff --git a/vllm/model_executor/layers/mamba/mamba_mixer2.py b/vllm/model_executor/layers/mamba/mamba_mixer2.py index 5bc890d007..1496515e77 100644 --- a/vllm/model_executor/layers/mamba/mamba_mixer2.py +++ b/vllm/model_executor/layers/mamba/mamba_mixer2.py @@ -180,6 +180,13 @@ class Mixer2RMSNormGated(CustomOp): norm_before_gate=False, ) + def forward_xpu( + self, + x: torch.Tensor, + gate: torch.Tensor, + ) -> torch.Tensor | tuple[torch.Tensor, torch.Tensor]: + return self.forward_cuda(x, gate) + def mamba_v2_sharded_weight_loader( shard_spec: list[tuple[int, int, float]], diff --git a/vllm/third_party/flash_linear_attention/ops/kda.py b/vllm/third_party/flash_linear_attention/ops/kda.py index 1701cd47ac..548f600812 100644 --- a/vllm/third_party/flash_linear_attention/ops/kda.py +++ b/vllm/third_party/flash_linear_attention/ops/kda.py @@ -529,6 +529,22 @@ class FusedRMSNormGated(CustomOp): residual_in_fp32=residual_in_fp32, ) + def forward_xpu( + self, + x: torch.Tensor, + g: torch.Tensor, + residual: torch.Tensor | None = None, + prenorm: bool = False, + residual_in_fp32: bool = False, + ) -> torch.Tensor: + return self.forward_cuda( + x, + g, + residual=residual, + prenorm=prenorm, + residual_in_fp32=residual_in_fp32, + ) + @triton.heuristics({"IS_VARLEN": lambda args: args["cu_seqlens"] is not None}) @triton.autotune(