[ROCm] Fix fused_moe_fake signature mismatch and other AITER bugs (#36100)
Signed-off-by: Li <chuali@amd.com>
This commit is contained in:
+5
-1
@@ -137,6 +137,10 @@ def _rocm_aiter_fused_moe_fake(
|
||||
a2_scale: torch.Tensor | None = None,
|
||||
num_local_tokens: torch.Tensor | None = None,
|
||||
output_dtype: torch.dtype | None = None,
|
||||
hidden_pad: int = 0,
|
||||
intermediate_pad: int = 0,
|
||||
bias1: torch.Tensor | None = None,
|
||||
bias2: torch.Tensor | None = None,
|
||||
) -> torch.Tensor:
|
||||
if output_dtype is not None:
|
||||
return torch.empty_like(hidden_states, dtype=output_dtype)
|
||||
@@ -1700,7 +1704,7 @@ class rocm_aiter_ops:
|
||||
)
|
||||
|
||||
@staticmethod
|
||||
def triton_fp4_gemm_dynamic_qaunt(
|
||||
def triton_fp4_gemm_dynamic_quant(
|
||||
x: torch.Tensor,
|
||||
weight: torch.Tensor,
|
||||
weight_scale: torch.Tensor,
|
||||
|
||||
Reference in New Issue
Block a user