!252 fix rnntloss label shape and grads

Merge pull request !252 from yanzhenxiang2020/fix_rnntloss_incubator

mindspore/ops/_grad/grad_nn_ops.py

@@ -521,11 +521,9 @@ def get_bprop_l2_loss(self):
 @bprop_getters.register(P.RNNTLoss)
 def get_bprop_rnnt_loss(self):
     """Grad definition for `RNNTLoss` operation."""
-    expand = P.ExpandDims()
 
     def bprop(acts, labels, act_lens, label_lens, out, dout):
-        grad_loss = out[1]
-        grad = grad_loss * expand(expand(expand(dout[0], -1), -1), -1)
+        grad = out[1]
         return grad, zeros_like(labels), zeros_like(act_lens), zeros_like(label_lens)
     return bprop
 

mindspore/ops/operations/nn_ops.py

@@ -1706,7 +1706,7 @@ class RNNTLoss(PrimitiveWithInfer):
 
     Inputs:
         - **acts** (Tensor[float32]) - Tensor of shape :math:`(B, T, U, V)`.
-        - **labels** (Tensor[int32]) - Tensor of shape :math:`(B, N)`.
+        - **labels** (Tensor[int32]) - Tensor of shape :math:`(B, U-1)`.
         - **input_lengths** (Tensor[int32]) - Tensor of shape :math:`(B,)`.
         - **label_lebgths** (Tensor[int32]) - Tensor of shape :math:`(B,)`.
 
@@ -1735,6 +1735,7 @@ class RNNTLoss(PrimitiveWithInfer):
         validator.check_integer('input_length_rank', len(input_length_shape), 1, Rel.EQ, self.name)
         validator.check_integer('label_length_rank', len(label_length_shape), 1, Rel.EQ, self.name)
         validator.check('labels shape[0]', labels_shape[0], 'acts shape[0]', acts_shape[0], Rel.EQ, self.name)
+        validator.check('labels shape[1]', labels_shape[1], 'acts shape[2]-1', acts_shape[2]-1, Rel.EQ, self.name)
         validator.check('input_length size', input_length_shape[0], 'acts shape[0]', acts_shape[0], Rel.EQ, self.name)
         validator.check('label_length size', label_length_shape[0], 'acts shape[0]', acts_shape[0], Rel.EQ, self.name)
         costs_shape = (acts_shape[0],)