Add missing yamls

Author: A-nnonymous

paddle/phi/ops/yaml/backward.yaml

@@ -2280,6 +2280,16 @@
   kernel :
     func : moe_combine_grad
 
+- backward_op : moe_gate_dispatch_grad
+  forward : moe_gate_dispatch (Tensor x, Tensor gate_logits, Tensor corr_bias, int64_t k, int64_t capacity, bool use_pad) -> Tensor(y), Tensor(combine_weights), Tensor(scatter_index), Tensor(expert_offset), Tensor(expert_id)
+  args : (Tensor combine_weights, Tensor scatter_index, Tensor expert_id, Tensor y_grad, Tensor combine_weights_grad, int64_t k, int64_t capacity, bool use_pad)
+  output : Tensor(x_grad), Tensor(gate_logits_grad)
+  infer_meta :
+    func : MoeGateDispatchGradInferMeta
+  kernel :
+    func : moe_gate_dispatch_grad
+    data_type : y_grad
+
 - backward_op : moe_gate_dispatch_partial_nosoftmaxtopk_grad
   forward : moe_gate_dispatch_partial_nosoftmaxtopk (Tensor x, Tensor combine_weights, Tensor expert_id, int64_t k, int64_t capacity, int64_t num_experts, bool use_pad, int64_t expert_start_index, int64_t expert_end_index, bool reverse_token_drop) -> Tensor(y), Tensor(combine_weights_out), Tensor(scatter_index), Tensor(scatter_index_rev), Tensor(expert_offset), Tensor(expert_nums_local)
   args : (Tensor combine_weights_out, Tensor scatter_index, Tensor scatter_index_rev, Tensor expert_offset, Tensor expert_nums_local, Tensor y_grad, Tensor combine_weights_out_grad, int64_t k, int64_t capacity, bool use_pad, int64_t expert_start_index, int64_t expert_end_index)

paddle/phi/ops/yaml/ops.yaml

@@ -3637,6 +3637,17 @@
     data_type : x
   backward : moe_combine_grad
 
+- op : moe_gate_dispatch
+  args : (Tensor x, Tensor gate_logits, Tensor corr_bias, int64_t k, int64_t capacity, bool use_pad)
+  output : Tensor(y), Tensor(combine_weights), Tensor(scatter_index), Tensor(expert_offset), Tensor(expert_id)
+  infer_meta :
+    func : MoeGateDispatchInferMeta
+  kernel :
+    func : moe_gate_dispatch
+    data_type : x
+  optional : corr_bias
+  backward : moe_gate_dispatch_grad
+  
 - op : moe_gate_dispatch_partial_nosoftmaxtopk
   args : (Tensor x, Tensor combine_weights, Tensor expert_id, int64_t k, int64_t capacity, int64_t num_experts, bool use_pad, int64_t expert_start_index, int64_t expert_end_index, bool reverse_token_drop)
   output : Tensor(y), Tensor(combine_weights_out), Tensor(scatter_index), Tensor(scatter_index_rev), Tensor(expert_offset), Tensor(expert_nums_local)