wehub-resource-sync
86 lines
3.5 KiB
C++
86 lines
3.5 KiB
C++
/* Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
|
|
|
|
Licensed under the Apache License, Version 2.0 (the "License");
|
|
you may not use this file except in compliance with the License.
|
|
You may obtain a copy of the License at
|
|
|
|
http://www.apache.org/licenses/LICENSE-2.0
|
|
|
|
Unless required by applicable law or agreed to in writing, software
|
|
distributed under the License is distributed on an "AS IS" BASIS,
|
|
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
See the License for the specific language governing permissions and
|
|
limitations under the License. */
|
|
|
|
#include "paddle/phi/infermeta/spmd_rules/label_smooth.h"
|
|
#include "glog/logging.h"
|
|
#include "paddle/phi/core/distributed/auto_parallel/dist_attr.h"
|
|
#include "paddle/phi/core/distributed/auto_parallel/dist_meta_tensor.h"
|
|
#include "paddle/phi/core/distributed/type_defs.h"
|
|
#include "paddle/phi/infermeta/spmd_rules/elementwise.h"
|
|
#include "paddle/phi/infermeta/spmd_rules/spmd_rule_macro_define.h"
|
|
#include "paddle/phi/infermeta/spmd_rules/utils.h"
|
|
|
|
namespace phi {
|
|
namespace distributed {
|
|
|
|
std::vector<int64_t> ReplicateWithExcludeAxes(
|
|
const std::vector<int64_t>& x, const std::vector<int64_t>& exclude_axes) {
|
|
std::vector<int64_t> res(x);
|
|
for (size_t i = 0; i < x.size(); ++i) {
|
|
if (res[i] == -1) {
|
|
continue;
|
|
}
|
|
|
|
if (std::find(exclude_axes.begin(), exclude_axes.end(), res[i]) !=
|
|
exclude_axes.end()) {
|
|
res[i] = -1;
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
SpmdInfo LabelSmoothInferSpmd(const DistMetaTensor& label,
|
|
const DistMetaTensor& prior_dist,
|
|
float epsilon) {
|
|
if (prior_dist.initialized()) {
|
|
const auto& label_dims_attr = label.dist_attr();
|
|
const auto& prior_dims_attr = prior_dist.dist_attr();
|
|
const auto& label_dims_mapping = label_dims_attr.dims_mapping();
|
|
const auto& prior_dims_mapping = prior_dims_attr.dims_mapping();
|
|
size_t label_rank = label.dims().size();
|
|
size_t prior_rank = prior_dist.dims().size();
|
|
|
|
if (label_rank > prior_rank) {
|
|
size_t extra_dims_count = label_rank - prior_rank;
|
|
std::vector<std::int64_t> extra_dims_mapping_values;
|
|
extra_dims_mapping_values.reserve(extra_dims_count);
|
|
for (size_t i = 0; i < extra_dims_count; ++i) {
|
|
extra_dims_mapping_values.push_back(label_dims_mapping[i]);
|
|
}
|
|
const auto& new_prior_dims_mapping = ReplicateWithExcludeAxes(
|
|
prior_dims_mapping, extra_dims_mapping_values);
|
|
TensorDistAttr new_prior_dist_attr = prior_dims_attr;
|
|
new_prior_dist_attr.set_dims_mapping(new_prior_dims_mapping);
|
|
DistMetaTensor modified_prior_dist(prior_dist.dims(),
|
|
new_prior_dist_attr);
|
|
VLOG(4) << "LabelSmoothInferSpmd call ElementwiseBinaryInferSpmd.";
|
|
return ElementwiseBinaryInferSpmd(label, modified_prior_dist);
|
|
}
|
|
VLOG(4) << "LabelSmoothInferSpmd call ElementwiseBinaryInferSpmd.";
|
|
return ElementwiseBinaryInferSpmd(label, prior_dist);
|
|
}
|
|
VLOG(4) << "LabelSmoothInferSpmd call ElementwiseUnaryInferSpmd.";
|
|
SpmdInfo unary_spmd_info = ElementwiseUnaryInferSpmd(label);
|
|
unary_spmd_info.first.push_back(TensorDistAttr());
|
|
return unary_spmd_info;
|
|
}
|
|
|
|
SpmdInfo LabelSmoothGradInferSpmd(const DistMetaTensor& out_grad,
|
|
float epsilon) {
|
|
VLOG(4) << "LabelSmoothGradInferSpmd call ElementwiseUnaryGradInferSpmd.";
|
|
return ElementwiseUnaryGradInferSpmd(out_grad);
|
|
}
|
|
} // namespace distributed
|
|
} // namespace phi
|