[BugFix] BF16 MoE Cutlass Backend Support EP (#5242)

Author: ckl117

fastdeploy/config.py

@@ -304,6 +304,8 @@ def override_name_from_config(self):
 
         if hasattr(self, "num_experts") and getattr(self, "moe_num_experts") is None:
             self.moe_num_experts = self.num_experts
+        if hasattr(self, "n_routed_experts") and getattr(self, "moe_num_experts") is None:
+            self.moe_num_experts = self.n_routed_experts
 
     def read_from_env(self):
         """

fastdeploy/model_executor/layers/moe/fused_moe_cutlass_backend.py

@@ -206,7 +206,10 @@ def apply_ep_prefill(
             tmp_ffn_out = recv_x
 
         # 4. EP combine
-        return self.ep_prefill_runner.combine(tmp_ffn_out, handle, recv_topk_weights)
+        tmp_ffn_out, event = self.ep_prefill_runner.combine(tmp_ffn_out, handle, recv_topk_weights)
+        if self.ep_prefill_runner.ep_engine.async_finish:
+            event.current_stream_wait()
+        return tmp_ffn_out
 
     def apply_ep_decode(
         self,
@@ -242,7 +245,7 @@ def apply_ep_decode(
         if self.moe_quant_type == "w4a8" or self.moe_quant_type == "w4afp8":
             num_local_experts, max_num, _ = permute_input.shape
             expert_idx_per_token = paddle.arange(num_local_experts)[:, None].tile([1, max_num])
-        elif self.moe_quant_type in ["weight_only_int8", "weight_only_int4"]:
+        elif self.moe_quant_type in ["weight_only_int8", "weight_only_int4", "w16a16"]:
             expert_idx_per_token = None
         else:
             raise NotImplementedError

fastdeploy/model_executor/layers/moe/fused_moe_triton_backend.py

@@ -808,7 +808,7 @@ def apply(
             N=hidden_size,
             K=moe_intermediate_size,
             stride_am=x_q.strides[0],
-            stride_ak=x_scale.strides[1],
+            stride_ak=x_q.strides[1],
             stride_be=layer.down_proj_weight.strides[0],
             stride_bk=layer.down_proj_weight.strides[2],
             stride_bn=layer.down_proj_weight.strides[1],

fastdeploy/model_executor/models/glm4_moe.py

@@ -494,6 +494,20 @@ def compute_logits(self, hidden_states: paddle.Tensor):
 
         return logits
 
+    def empty_input_forward(self):
+        """
+        empty_input_forward
+        """
+        fake_hidden_states = paddle.ones(
+            shape=[1, self.fd_config.model_config.hidden_size],
+            dtype=paddle.get_default_dtype(),
+        )
+        for i in range(
+            self.fd_config.model_config.first_k_dense_replace,
+            self.fd_config.model_config.num_hidden_layers,
+        ):
+            self.model.layers[i].mlp.experts(fake_hidden_states, self.model.layers[i].mlp.gate)
+
     def forward(
         self,
         ids_remove_padding: paddle.Tensor,