mindspore-Ecosystem
/
mindspore
21
0
Fork 27
Code Issues Pull Requests 6 Projects Releases Wiki Activity

fix incorrect input shape check of TruncatedNormal operator.

This commit is contained in:
yangshuo 2022-12-08 16:53:17 +08:00
parent 813187cf04
commit 1a4d660356
4 changed files with 18 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
mindspore/ccsrc/plugin/device/cpu/kernel/truncated_normal_cpu_kernel.cc
View File

@ -64,6 +64,14 @@ int TruncatedNormalCPUKernelMod::Resize(const BaseOperatorPtr &base_operator,
if (ret != KRET_OK) {
return ret;
}
auto shape_input = inputs[kIndex0]->GetShapeVector();
if (shape_input.size() != kInputDims) {
MS_EXCEPTION(ValueError) << "For '" << kernel_name_ << "', The input tensor must be a 1-D tensor.";
}
if (shape_input[kIndex0] < kInputSizes) {
MS_EXCEPTION(ValueError) << "For '" << kernel_name_ << "', the input tensor shape must >= 2, but got "
<< shape_input[kIndex0];
}
input_type_ = inputs[kIndex0]->GetDtype();
output_type_ = outputs[kIndex0]->GetDtype();
return KRET_OK;
@ -126,7 +134,7 @@ std::vector<std::pair<KernelAttr, TruncatedNormalCPUKernelMod::TruncatedNormalFu
{KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeFloat32),
&TruncatedNormalCPUKernelMod::LaunchKernel<int32_t, float, float>},
{KernelAttr().AddInputAttr(kNumberTypeInt64).AddOutputAttr(kNumberTypeFloat64),
&TruncatedNormalCPUKernelMod::LaunchKernel<int32_t, double, double>},
&TruncatedNormalCPUKernelMod::LaunchKernel<int64_t, double, double>},
{KernelAttr().AddInputAttr(kNumberTypeInt64).AddOutputAttr(kNumberTypeFloat16),
&TruncatedNormalCPUKernelMod::LaunchKernel<int64_t, float16, float>},
{KernelAttr().AddInputAttr(kNumberTypeInt64).AddOutputAttr(kNumberTypeFloat32),

10
mindspore/core/ops/reduce_scatter.cc
View File

@ -58,8 +58,8 @@ class ReduceScatterInfer : public abstract::OpInferBase {
MS_ERROR_IF_NULL_W_RET_VAL(value_ptr, std::make_shared<abstract::Shape>());
auto rank_size = static_cast<int>(GetValue<int64_t>(value_ptr));
if (rank_size == 0) {
MS_LOG(ERROR) << "For '" << primitive->name() << "', the 'rank_size' can not be zero, but got " << rank_size;
return std::make_shared<abstract::Shape>();
MS_EXCEPTION(ValueError) << "For '" << primitive->name() << "', the 'rank_size' can not be zero, but got "
<< rank_size;
}
auto abstract_shape = input_args[kIndex0]->BuildShape();
MS_ERROR_IF_NULL_W_RET_VAL(abstract_shape, std::make_shared<abstract::Shape>());
@ -68,9 +68,9 @@ class ReduceScatterInfer : public abstract::OpInferBase {
}
auto shape = CheckAndConvertUtils::ConvertShapePtrToShapeMap(abstract_shape)[kShape];
if (shape.empty() || shape[0] % rank_size != 0) {
MS_LOG(ERROR) << "the first dimension for 'input_shape' must be divided by 'rank_size', but got input_shape[0]: "
<< shape[0] << ", rank_size: " << rank_size;
return std::make_shared<abstract::Shape>();
MS_EXCEPTION(ValueError)
<< "the first dimension for 'input_shape' must be divided by 'rank_size', but got input_shape[0]: " << shape[0]
<< ", rank_size: " << rank_size;
}
auto out_shape = shape;
out_shape[0] = static_cast<int64_t>(shape[0] / rank_size);

7
mindspore/core/ops/truncated_normal.cc
View File

@ -39,13 +39,6 @@ abstract::ShapePtr TruncatedNormalInferShape(const PrimitivePtr &primitive,
if (IsDynamicRank(shape_input)) {
return std::make_shared<abstract::Shape>(std::vector<int64_t>{abstract::Shape::kShapeRankAny});
}
if (shape_input.size() != kInputDims) {
MS_EXCEPTION(ValueError) << "For '" << primitive->name() << "', The input tensor must be a 1-D tensor.";
}
if (shape_input[kInputIndex0] < kInputSizes) {
MS_EXCEPTION(ValueError) << "For '" << primitive->name() << "', the input tensor shape must >= 2, but got "
<< shape_input[kInputIndex0];
}
MS_EXCEPTION_IF_NULL(primitive);
const uint32_t kInpuDims = 1;
auto max_length_ptr = primitive->GetAttr("max_length");

5
mindspore/python/mindspore/ops/_grad_experimental/grad_nn_ops.py
View File

@ -147,7 +147,10 @@ def get_bprop_pad_v3(self):
def bprop(x, paddings, constant_values, out, dout):
if mode == 'constant':
neg_paddings = tuple(-x for x in paddings)
if isinstance(paddings, (list, tuple)):
neg_paddings = tuple(-x for x in paddings)
else:
neg_paddings = -paddings
dx = pad_v3_grad(dout, neg_paddings, zeros_like(constant_values))
else:
dx = pad_v3_grad(dout, paddings)