[NPU] bugfix for Base class add mamba_track_indices parameter (#29999)

python/sglang/srt/hardware_backend/npu/attention/ascend_hybrid_linear_attn_backend.py

@@ -132,6 +132,8 @@ class AscendMambaAttnBackendBase(MambaAttnBackendBase):
         seq_lens_cpu: Optional[torch.Tensor],
         num_padding: Optional[int] = None,
         in_capture: bool = False,
+        *args,
+        **kwargs,
     ):
         # out_graph passes seq_lens_cpu=None at capture; mirror the base guard.
         if seq_lens_cpu is None:

python/sglang/srt/hardware_backend/npu/quantization/fused_moe_method_npu.py

@@ -284,7 +284,7 @@ def npu_fused_moe_without_routing_weights_bf16(
     # gmm1: gate_up_proj
     hidden_states = torch.ops.npu.npu_grouped_matmul(
         x=[hidden_states],
-        weight=[layer.w13_weight],
+        weight=[layer.w13_weight.transpose(1, 2)],
         split_item=2,
         group_list_type=group_list_type,
         group_type=0,
@@ -297,7 +297,7 @@ def npu_fused_moe_without_routing_weights_bf16(
     # gmm2: down_proj
     hidden_states = torch.ops.npu.npu_grouped_matmul(
         x=[hidden_states],
-        weight=[layer.w2_weight],
+        weight=[layer.w2_weight.transpose(1, 2)],
         split_item=2,
         group_list_type=group_list_type,
         group_type=0,