xxq250
/
mindspore
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[MSLITE][Develop] support al_bert inferance

This commit is contained in:
yangruoqi713 2021-03-17 16:48:32 +08:00
parent e8fb3dfcc9
commit 4e9002329f
3 changed files with 8 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
mindspore/lite/nnacl/infer/arithmetic_grad_infer.c
View File

@ -103,3 +103,4 @@ int ArithmeticGradInferShape(const TensorC *const *inputs, size_t inputs_size, T
REG_INFER(DivGrad, PrimType_DivGrad, ArithmeticGradInferShape)
REG_INFER(MulGrad, PrimType_MulGrad, ArithmeticGradInferShape)
REG_INFER(MinimumGrad, PrimType_MinimumGrad, ArithmeticGradInferShape)

9
mindspore/lite/nnacl/infer/layer_norm_infer.c
View File

@ -37,6 +37,8 @@ int LayerNormInferShape(const TensorC *const *inputs, size_t inputs_size, Tensor
if (!param->op_parameter_.infer_flag_) {
return NNACL_INFER_INVALID;
}
param->begin_norm_axis_ =
param->begin_norm_axis_ < 0 ? param->begin_norm_axis_ + input->shape_size_ : param->begin_norm_axis_;
SetShapeTensor(output, input);
// take care of other outputs
if (outputs_size == 3) {
@ -45,10 +47,9 @@ int LayerNormInferShape(const TensorC *const *inputs, size_t inputs_size, Tensor
SetDataTypeFormat(output_mean, input);
SetDataTypeFormat(output_var, input);
int size = 0;
for (int i = param->begin_norm_axis_; i < input->shape_size_; i++) {
output_mean->shape_[size] = input->shape_[i];
output_var->shape_[size] = input->shape_[i];
size++;
for (; size < param->begin_norm_axis_; size++) {
output_mean->shape_[size] = input->shape_[size];
output_var->shape_[size] = input->shape_[size];
}
output_mean->shape_size_ = size;
output_var->shape_size_ = size;

4
mindspore/lite/src/runtime/kernel/arm/fp32_grad/activation_grad.cc
View File

@ -35,8 +35,8 @@ using mindspore::schema::PrimitiveType_ActivationGrad;
namespace mindspore::kernel {
int ActivationGradCPUKernel::Init() {
if (in_tensors_.size() != 2) {
MS_LOG(ERROR) << "ActivationGrad should have 2 input tensors";
if (in_tensors_.size() < 2) {
MS_LOG(ERROR) << "ActivationGrad should have more than 2 input tensors";
return RET_ERROR;
}
return RET_OK;