[Hardware][Gaudi][BugFix] fix arguments of hpu fused moe (#15945)

Signed-off-by: zhenwei <[email protected]>

Author: zhenwei-intel

vllm/model_executor/layers/fused_moe/layer.py

@@ -254,9 +254,12 @@ def forward_hpu(
         renormalize: bool,
         topk_group: Optional[int] = None,
         num_expert_group: Optional[int] = None,
+        global_num_experts: int = -1,
+        expert_map: Optional[torch.Tensor] = None,
         custom_routing_function: Optional[Callable] = None,
         scoring_func: str = "softmax",
-        e_score_correction_bias: Optional[torch.Tensor] = None
+        e_score_correction_bias: Optional[torch.Tensor] = None,
+        activation: str = "silu",
     ) -> torch.Tensor:
         assert not use_grouped_topk
         assert num_expert_group is None
@@ -472,7 +475,7 @@ def __init__(
                              "non-grouped topk.")
         if current_platform.is_hpu():
             from vllm_hpu_extension.ops import DynamicFusedMOE
-            self.hpu_fused_moe = DynamicFusedMOE(self.num_experts)
+            self.hpu_fused_moe = DynamicFusedMOE(self.global_num_experts)
 
         # Note: get_quant_method will look at the layer's local_num_experts
         # for heuristic purposes, so it must be initialized first.