[Feat] ernie4_5_vl_moe support CudaGraph (#3226)

* delete dynamic control flow for decode

* coda-style

* fix scatter/gather typos and use input stream instead default stream

* support 0-Size Tensor

* update runner and model

* using static mem address as input

* fix mem leak

* refine code

* update mm_buffer

* fix typo

* fix buffersize

* fix unk token

* refine code

* refine

* support other arch

* open cudagraph in vlci

* fix

* update

* update

* update

* fix cmd

* update

---------

Co-authored-by: aquagull <hongyuh@qq.com>
Co-authored-by: Yuanle Liu <yuanlehome@163.com>

custom_ops/gpu_ops/moe/moe_dispatch.cu

@@ -36,6 +36,9 @@ void MoeDispatchKernel(
     paddle::Tensor *topk_idx, paddle::Tensor *expert_idx_per_token) {
   using namespace phi;
 
+  if (num_rows == 0){
+    return;
+  }
   typedef PDTraits<T> traits_;
   typedef typename traits_::DataType DataType_;
   typedef typename traits_::data_t data_t;
@@ -185,6 +188,15 @@ std::vector<paddle::Tensor> MoeExpertDispatch(
   auto expert_idx_per_token =
       GetEmptyTensor({num_rows * moe_topk}, paddle::DataType::INT32, place);
 
+  if (token_rows == 0){
+    return {permute_input,
+            tokens_expert_prefix_sum,
+            permute_indices_per_token,
+            topk_weight,
+            topk_idx,
+            expert_idx_per_token};
+  }
+
   switch (input_type) {
   case paddle::DataType::BFLOAT16:
     MoeDispatchKernel<paddle::DataType::BFLOAT16>(

custom_ops/gpu_ops/moe/moe_ffn.cu

@@ -412,7 +412,9 @@ const auto t_type = (quant_method == "w4a8") ? up_gate_proj_scale.get().dtype()
                     (quant_method == "w4afp8") ? paddle::DataType::BFLOAT16 :
                     permute_input.dtype();
     auto ffn_out = paddle::empty_like(permute_input, t_type);
-
+    if(permute_input.numel() == 0){
+        return ffn_out;
+    }
     switch (t_type) {
         case paddle::DataType::BFLOAT16:
             MoeFFNKernel<paddle::DataType::BFLOAT16>(permute_input,

custom_ops/gpu_ops/moe/moe_reduce.cu

@@ -59,6 +59,10 @@ paddle::Tensor MoeExpertReduceFunc(
 
   auto output = GetEmptyTensor({num_rows, hidden_size}, input_type, place);
 
+  if(num_rows == 0){
+    return output;
+  }
+
   switch (input_type) {
   case paddle::DataType::BFLOAT16:
     MoeReduceKernel<paddle::DataType::BFLOAT16>(