[Model][Bugfix] fix ernie45 vl run failed from shared experts optimization (vllm-project#26885)

Signed-off-by: wangyafeng <wangyafeng@baidu.com>
Signed-off-by: Alberto Perdomo <aperdomo@redhat.com>

Author: CSWYF3634076

vllm/model_executor/models/ernie45_vl_moe.py

@@ -341,7 +341,10 @@ def forward(
             # text and vision modals input
             visual_token_mask = visual_token_mask.repeat(1, self.hidden_size).bool()
             text_token_mask = ~visual_token_mask
-            final_hidden_states = torch.zeros_like(hidden_states)
+            final_experts_hidden_states = torch.zeros_like(hidden_states)
+            final_shared_ouput = (
+                torch.zeros_like(hidden_states) if self.has_shared_experts else None
+            )
 
             text_hidden_states = hidden_states[text_token_mask].reshape(
                 -1, self.hidden_size
@@ -353,16 +356,26 @@ def forward(
             text_router_logits, _ = self.text_experts_gate(
                 text_hidden_states.to(dtype=torch.float32)
             )
-            final_hidden_states[text_token_mask] = self.text_experts(
+            text_shared_ouput, text_experts_output = self.text_experts(
                 hidden_states=text_hidden_states, router_logits=text_router_logits
-            ).flatten()
+            )
+            final_experts_hidden_states[text_token_mask] = text_experts_output.flatten()
+            if self.has_shared_experts:
+                final_shared_ouput[text_token_mask] = text_shared_ouput.flatten()
 
             vision_router_logits, _ = self.vision_experts_gate(
                 vision_hidden_states.to(dtype=torch.float32)
             )
-            final_hidden_states[visual_token_mask] = self.vision_experts(
+            vision_shared_ouput, vision_experts_output = self.vision_experts(
                 hidden_states=vision_hidden_states, router_logits=vision_router_logits
-            ).flatten()
+            )
+            final_experts_hidden_states[visual_token_mask] = (
+                vision_experts_output.flatten()
+            )
+            if self.has_shared_experts:
+                final_shared_ouput[visual_token_mask] = vision_shared_ouput.flatten()
+
+            final_hidden_states = (final_shared_ouput, final_experts_hidden_states)
         else:
             # only text modal input
             text_router_logits, _ = self.text_experts_gate(
@@ -374,7 +387,11 @@ def forward(
             )
 
         if self.has_shared_experts:
+            # for shared_experts model
             final_hidden_states = final_hidden_states[0] + final_hidden_states[1]
+        else:
+            # for not shared_experts model
+            final_hidden_states = final_hidden_states[1]
 
         if self.tp_size > 1:
             final_hidden_states = (