fix ut

Signed-off-by: 白永斌 <[email protected]>
Signed-off-by: 欧派果奶我还要 <[email protected]>

Author: 白永斌

tests/ut/ops/test_fused_moe.py

@@ -409,9 +409,9 @@ def test_unified_apply_mlp_without_quantization(self,
         topk_scales = torch.randn(10, 1, dtype=torch.float16)
 
         result = unified_apply_mlp(hidden_states=hidden_states,
-                                   w1=w1,
+                                   w1=[w1],
                                    w1_scale=None,
-                                   w2=w2,
+                                   w2=[w2],
                                    w2_scale=None,
                                    group_list=group_list,
                                    dynamic_scale=None,
@@ -512,9 +512,9 @@ def test_unified_apply_mlp_without_quantization_310p(
         topk_scales = torch.randn(10, 1, dtype=torch.float16)
 
         result = unified_apply_mlp(hidden_states=hidden_states,
-                                   w1=w1,
+                                   w1=[w1],
                                    w1_scale=None,
-                                   w2=w2,
+                                   w2=[w2],
                                    w2_scale=None,
                                    group_list=group_list,
                                    dynamic_scale=None,

vllm_ascend/ops/fused_moe/moe_comm_method.py

@@ -137,7 +137,6 @@ def fused_experts(
         permuted_hidden_states, expert_tokens, dynamic_scale, group_list_type, topk_scales, context_metadata = \
             results["hidden_states"], results["group_list"], results.get("dynamic_scale"), results["group_list_type"], results.get("topk_scales"), results.get("context_metadata")
 
-        assert w1_scale is not None and w2_scale is not None
         mlp_output = unified_apply_mlp(hidden_states=permuted_hidden_states,
                                        w1=w1,
                                        w1_scale=w1_scale,

vllm_ascend/ops/fused_moe/moe_mlp.py

@@ -262,9 +262,7 @@ def unquant_apply_mlp(hidden_states: torch.Tensor,
 
 def unified_apply_mlp(hidden_states: torch.Tensor,
                       w1: list[torch.Tensor],
-                      w1_scale: list[torch.Tensor],
                       w2: list[torch.Tensor],
-                      w2_scale: list[torch.Tensor],
                       group_list: torch.Tensor,
                       w1_scale: Optional[list[torch.Tensor]] = None,
                       w2_scale: Optional[list[torch.Tensor]] = None,