[FEAT] [ROCm] Upgrade AITER Fused MoE kernels. (vllm-project#18271)

Signed-off-by: vllmellm <[email protected]>
Signed-off-by: amit <[email protected]>

Author: vllmellm

vllm/model_executor/layers/fused_moe/layer.py

@@ -419,10 +419,8 @@ def process_weights_after_loading(self, layer: torch.nn.Module) -> None:
             shuffle_weights)
 
         if self.rocm_aiter_moe_enabled:
-            # use 2stage ck moe layout
-            shuffled_w13, shuffled_w2 = shuffle_weights(layer.w13_weight.data,
-                                                        layer.w2_weight.data,
-                                                        layout=(32, 32))
+            shuffled_w13, shuffled_w2 = shuffle_weights(
+                layer.w13_weight.data, layer.w2_weight.data)
 
             layer.w13_weight.data = shuffled_w13
             layer.w2_weight.data = shuffled_w2