fix missing param

Author: Wanglongzhi2001

fastdeploy/model_executor/layers/moe/moe.py

@@ -573,7 +573,7 @@ def load_state_dict(self, state_dict, is_rearrange: bool = False):
         else:
             self.quant_method.process_loaded_weights(self, state_dict)
 
-    def forward_split_allgather(self, x: paddle.Tensor, gate: nn.Layer):
+    def forward_split_allgather(self, x: paddle.Tensor, gate: nn.Layer, forward_meta: ForwardMeta):
         """
         Forward split allgather function.
         """
@@ -588,7 +588,7 @@ def forward_split_allgather(self, x: paddle.Tensor, gate: nn.Layer):
         if end_offset > token_num:
             end_offset = token_num
         part_x[: (end_offset - start_offset), :] = x[start_offset:end_offset, :]
-        out = self.quant_method.apply(self, part_x, gate)
+        out = self.quant_method.apply(self, part_x, gate, forward_meta)
         multi_outs = paddle.zeros([token_num_per_rank * self.tp_size, x.shape[1]], dtype=x.dtype)
         paddle.distributed.all_gather(multi_outs, out, self.tp_group)
         out = multi_outs[:token_num, :]

fastdeploy/model_executor/models/deepseek_v3.py

@@ -189,8 +189,8 @@ def load_state_dict(self, state_dict):
 
     def forward(self, hidden_states: paddle.Tensor, forward_meta: ForwardMeta):
         """ """
-        shared_experts_out = self.shared_experts(hidden_states)
-        moe_out = self.experts(hidden_states, self.gate)
+        shared_experts_out = self.shared_experts(hidden_states, forward_meta)
+        moe_out = self.experts(hidden_states, self.gate, forward_meta)
         moe_out = moe_out + shared_experts_out
         # We do to TP all reduce after the sum of experts.
         if self.tp_size > 1:

fastdeploy/model_executor/models/ernie4_5_moe.py

@@ -224,7 +224,7 @@ def forward(
             forward_meta=forward_meta,
         )
         if self.num_shared_experts > 0:
-            s_x = self.shared_experts(hidden_states)
+            s_x = self.shared_experts(hidden_states, forward_meta)
             out = out + s_x
         return out
 
@@ -353,8 +353,8 @@ def forward(
         )
 
         hidden_states = self.mlp(
-            forward_meta=forward_meta,
             hidden_states=hidden_states,
+            forward_meta=forward_meta,
         )
 
         return hidden_states, residual

fastdeploy/model_executor/models/ernie4_5_vl/ernie4_5_vl_moe.py

@@ -271,7 +271,7 @@ def load_state_dict(self, state_dict):
 
     def forward(self, hidden_states: paddle.Tensor, forward_meta: ForwardMeta, vl_moe_meta: VLMoEMeta):
         if self.num_shared_experts > 0:
-            shared_experts_out = self.shared_experts(hidden_states)
+            shared_experts_out = self.shared_experts(hidden_states, forward_meta)
         hidden_states, text_input, image_input = text_image_gather_scatter(
             hidden_states,
             vl_moe_meta.text_input,