未验证 提交 c167a4b4 编写于 作者: F fuyinno4 提交者: GitHub

Fix shrink-dense and add scale-datanorm (#18746)

Fix FleetWrapper:
1. fix shrink dense: just scale show
2. add datanorm scale: divide datanorm's gradient by batch_size
上级 d3ac561d
develop 2.0.1-rocm-post Ligoml-patch-1 OliverLPH-patch-1 OliverLPH-patch-2 PaddlePM-patch-1 PaddlePM-patch-2 ZHUI-patch-1 add_default_att add_model_benchmark_ci add_some_yaml_config addfile all_new_design_exec ascendrc ascendrelease cherry_undefined_var compile_windows cp_2.4_fix_numpy delete_2.0.1-rocm-post delete_add_default_att delete_all_new_design_exec delete_ascendrc delete_compile_windows delete_delete_addfile delete_disable_iterable_dataset_unittest delete_fix_dataloader_memory_leak delete_fix_imperative_dygraph_error delete_fix_retry_ci delete_fix_undefined_var delete_improve_sccache delete_paddle_tiny_install delete_paralleltest delete_prv-disable-more-cache delete_revert-31068-fix_conv3d_windows delete_revert-31562-mean delete_revert-33630-bug-fix delete_revert-34159-add_npu_bce_logical_dev delete_revert-34910-spinlocks_for_allocator delete_revert-35069-revert-34910-spinlocks_for_allocator delete_revert-36057-dev/read_flags_in_ut dingjiaweiww-patch-1 disable_iterable_dataset_unittest dy2static enable_eager_model_test final_state_gen_python_c final_state_intermediate fix-numpy-issue fix_concat_slice fix_dataloader_memory_leak fix_dlpack_for fix_imperative_dygraph_error fix_npu_ci fix_op_flops fix_retry_ci fix_rnn_docs fix_tensor_type fix_undefined_var fix_var_stop_gradient_error fixiscan fixiscan1 fixiscan2 fixiscan3 github/fork/123malin/netifaces github/fork/123malin/tdm_abacus github/fork/AshburnLee/dev_unique github/fork/ForFishes/fix_memory_matmul github/fork/ForFishes/rm_fluid github/fork/LielinJiang/move-2.0-api github/fork/LielinJiang/visual-dl-cb github/fork/LiuChiachi/add-transformer-generate-square-subsequent-mask-api github/fork/LiuChiachi/fix-example-code-for-hapi-Model github/fork/LiuChiachi/remove-input-requirment-in-dygraph-Model github/fork/MrChengmo/fix_ps_profiler github/fork/MrChengmo/update_ps_heter github/fork/PWhiddy/patch-1 github/fork/Shixiaowei02/dev/save_load_upgrade github/fork/TCChenlong/fix_hapi github/fork/TCChenlong/fix_inden github/fork/Thunderbrook/xpu_slice github/fork/XieYunshen/disable_ut_test_parallel_executor_fetch_isolated_var github/fork/XieYunshen/disable_ut_test_parallel_executor_fetch_isolated_var_2 github/fork/XieYunshen/disable_ut_test_parallel_executor_fetch_isolated_var_3 github/fork/XieYunshen/timeout_20S_ut github/fork/ZeyuChen/remove-nltk github/fork/arlesniak/arlesniak/selective__mkldnn_flags github/fork/baiyfbupt/code_doc_mig github/fork/chalsliu/set_timeout github/fork/chen-zhiyu/develop github/fork/chenwhql/ci/try_to_find_test_buffer_shared_memory_reuse_pass_error github/fork/chenwhql/dygraph/remove_scale_loss_and_apply_collective_grads github/fork/chenwhql/saveload/add_get_inference_program github/fork/chenwhql/saveload/remove_save_load_config github/fork/cryoco/pass-compatibility-trt github/fork/danleifeng/isempty_api2.0 github/fork/frankwhzhang/api_transfer github/fork/hbwx24/error_msg/cuda_kernel_error_msg github/fork/heavengate/cherry_yolo_box github/fork/heavengate/update_yolo_box github/fork/iclementine/rnn_fix github/fork/iducn/testestse github/fork/jczaja/prv-25537-fix github/fork/jeff41404/release/1.8 github/fork/jiweibo/api_2.0 github/fork/jiweibo/fix_lite_resnet50_test github/fork/juncaipeng/fix_doc_1 github/fork/lfchener/sample_code github/fork/littletomatodonkey/fix_reg_doc github/fork/liym27/dy2stat_update_assign_to_rc20 github/fork/luotao1/profiler_ut github/fork/mapingshuo/add_wait github/fork/mapingshuo/doc_2.0 github/fork/mapingshuo/zero-0.5 github/fork/miraiwk/dev github/fork/pangyoki/add-Categorical-class-branch github/fork/pangyoki/add-multinomial-op-branch github/fork/pangyoki/fix-test_distritbution-CI github/fork/qjing666/doublegrad github/fork/qjing666/fix_hdfs_download github/fork/sandyhouse/add_gather_etc github/fork/sandyhouse/add_send_recv_alltoall_etc github/fork/sandyhouse/pipeline_exe_run github/fork/seiriosPlus/feature/large_scale_kv_save_delta github/fork/seiriosPlus/fix/paddle_errors_fix github/fork/seiriosPlus/fix/paddle_op_errors github/fork/shangzhizhou/fix_test_activation_op_random_bug github/fork/smallv0221/yxp0924 github/fork/smallv0221/yxp0925 github/fork/swtkiwi/del-matplotlib github/fork/tianshuo78520a/kunlun_test github/fork/tianshuo78520a/update_dockerfile github/fork/wanghaoshuang/bert_fuse github/fork/wanghaoshuang/label_smooth github/fork/wanghuancoder/develop_CUDASynchronize github/fork/wanghuancoder/develop_Layer_doc github/fork/wanghuancoder/develop_ParameterList_doc github/fork/wanghuancoder/develop_Sequential_doc github/fork/wanghuancoder/develop_bilinear_tensor_product github/fork/wanghuancoder/develop_coverage_build_sh github/fork/wanghuancoder/develop_in_dynamic_mode_doc github/fork/wanghuancoder/develop_unique_name_doc github/fork/wangxicoding/fleet_meta_combine github/fork/wawltor/error_message_fix_5 github/fork/willthefrog/remove_l2_norm github/fork/windstamp/momentum_op github/fork/windstamp/mv_op_5 github/fork/windstamp/normal_api github/fork/wojtuss/wojtuss/fusion_gru_quantization github/fork/wojtuss/wojtuss/quantization-with-shift github/fork/wzzju/fix_err_info github/fork/wzzju/pure_fp16 github/fork/xiemoyuan/op_error_message github/fork/xiemoyuan/optimize_error_message github/fork/yaoxuefeng6/fix_doc github/fork/yaoxuefeng6/mod_dataset_v2 github/fork/yongqiangma/lod github/fork/ysh329/fix-clip-by-norm-error github/fork/ysh329/fix-error-clip-by-value github/fork/yukavio/error_info github/fork/zhangting2020/conv_filter_grad github/fork/zhangting2020/is_compile_with_cuda github/fork/zhangting2020/place_doc github/fork/zhangting2020/program github/fork/zhhsplendid/fix_any github/fork/zhhsplendid/refine_api2 github/fork/zhhsplendid/refine_api2_test github/fork/zhhsplendid/refine_api_test_ptb_lm github/fork/zhhsplendid/refine_api_test_resnet github/fork/zhhsplendid/refine_api_test_simnet github/fork/zhiqiu/dev/refine_initializer github/fork/zhiqiu/dev/remove_inplace_argument github/fork/zlsh80826/nvinfer_plugin_var_len_cuda11 improve_sccache incubate/frl_train_eval incubate/infrt inplace_addto layer_norm make_flag_adding_easier matmul_double_grad move_embedding_to_phi move_histogram_to_pten move_sgd_to_phi move_slice_to_pten move_temporal_shift_to_phi move_yolo_box_to_phi npu_fix_alloc numel paddle_tiny_install paralleltest preln_ernie prv-disable-more-cache prv-md-even-more prv-onednn-2.5 prv-reshape-mkldnn-ut2 pten_tensor_refactor release/1.6 release/1.7 release/1.8 release/2.0 release/2.0-alpha release/2.0-beta release/2.0-rc release/2.0-rc1 release/2.1 release/2.2 release/2.3 release/2.3-fc-ernie-fix release/2.4 revert-24981-add_device_attr_for_regulization revert-26856-strategy_example2 revert-27520-disable_pr revert-31068-fix_conv3d_windows revert-31562-mean revert-32290-develop-hardlabel revert-33037-forci revert-33475-fix_cifar_label_dimension revert-33630-bug-fix revert-34159-add_npu_bce_logical_dev revert-34406-add_copy_from_tensor revert-34910-spinlocks_for_allocator revert-35069-revert-34910-spinlocks_for_allocator revert-36057-dev/read_flags_in_ut revert-36201-refine_fast_threaded_ssa_graph_executor revert-36985-add_license revert-37318-refactor_dygraph_to_eager revert-37926-eager_coreops_500 revert-37956-revert-37727-pylayer_support_tuple revert-38100-mingdong revert-38301-allocation_rearrange_pr revert-38703-numpy_bf16_package_reupload revert-38732-remove_useless_header_in_elementwise_mul_grad revert-38959-Reduce_Grad revert-39143-adjust_empty revert-39227-move_trace_op_to_pten revert-39268-dev/remove_concat_fluid_kernel revert-40170-support_partial_grad revert-41056-revert-40727-move_some_activaion_to_phi revert-41065-revert-40993-mv_ele_floordiv_pow revert-41068-revert-40790-phi_new revert-41944-smaller_inference_api_test revert-42149-do-not-reset-default-stream-for-stream-safe-cuda-allocator revert-43155-fix_ut_tempfile revert-43882-revert-41944-smaller_inference_api_test revert-45808-phi/simplify_size_op revert-46827-deform_comment revert-47325-remove_cudnn_hardcode revert-47645-add_npu_storage_dims revert-48815-set_free_when_no_cache_hit_default_value_true revert-49654-prim_api_gen revert-49763-fix_static_composite_gen rocm_dev_0217 support-0D-sort support_weight_transpose test_benchmark_ci test_feature_precision_test_c test_for_Filtetfiles test_model_benchmark test_model_benchmark_ci zhiqiu-patch-1 v2.4.1 v2.4.0 v2.4.0-rc0 v2.3.2 v2.3.1 v2.3.0 v2.3.0-rc0 v2.2.2 v2.2.1 v2.2.0 v2.2.0-rc0 v2.2.0-bak0 v2.1.3 v2.1.2 v2.1.1 v2.1.0 v2.1.0-rc0 v2.0.2 v2.0.1 v2.0.0 v2.0.0-rc1 v2.0.0-rc0 v2.0.0-beta0 v2.0.0-alpha0 v1.8.5 v1.8.4 v1.8.3 v1.8.2 v1.8.1 v1.8.0 v1.7.2 v1.7.1 v1.7.0 v1.6.3 v1.6.2 v1.6.1 v1.6.0 v1.6.0-rc0
无相关合并请求
......@@ -182,6 +182,7 @@ class DownpourWorker : public HogwildWorker {
bool dump_slot_;
bool need_to_push_sparse_;
DownpourWorkerParameter param_;
float scale_datanorm_;
// just save the value in param_ for easy access
std::map<uint64_t, std::string> label_var_name_;
std::map<uint64_t, std::vector<std::string>> sparse_key_names_;
......
......@@ -64,6 +64,7 @@ void DownpourWorker::Initialize(const TrainerDesc& desc) {
fleet_ptr_ = FleetWrapper::GetInstance();
fetch_config_ = desc.fetch_config();
use_cvm_ = desc.use_cvm();
scale_datanorm_ = desc.scale_datanorm();
dump_slot_ = desc.dump_slot();
}
......@@ -298,7 +299,8 @@ void DownpourWorker::TrainFilesWithProfiler() {
uint64_t tid = static_cast<uint64_t>(
param_.program_config(0).push_dense_table_id(i));
fleet_ptr_->PushDenseVarsAsync(
*thread_scope_, tid, dense_grad_names_[tid], &push_sparse_status_);
*thread_scope_, tid, dense_grad_names_[tid], &push_sparse_status_,
scale_datanorm_, cur_batch);
}
timeline.Pause();
push_dense_time += timeline.ElapsedSec();
......@@ -467,7 +469,8 @@ void DownpourWorker::TrainFiles() {
uint64_t tid = static_cast<uint64_t>(
param_.program_config(0).push_dense_table_id(i));
fleet_ptr_->PushDenseVarsAsync(
*thread_scope_, tid, dense_grad_names_[tid], &push_sparse_status_);
*thread_scope_, tid, dense_grad_names_[tid], &push_sparse_status_,
scale_datanorm_, cur_batch);
}
VLOG(3) << "push dense gradient done.";
......
......@@ -264,7 +264,8 @@ void FleetWrapper::PushDenseVarsSync(
void FleetWrapper::PushDenseVarsAsync(
const Scope& scope, const uint64_t table_id,
const std::vector<std::string>& var_names,
std::vector<::std::future<int32_t>>* push_sparse_status) {
std::vector<::std::future<int32_t>>* push_sparse_status,
float scale_datanorm, int batch_size) {
#ifdef PADDLE_WITH_PSLIB
std::vector<paddle::ps::Region> regions;
for (auto& t : var_names) {
......@@ -272,6 +273,20 @@ void FleetWrapper::PushDenseVarsAsync(
LoDTensor* tensor = var->GetMutable<LoDTensor>();
int count = tensor->numel();
float* g = tensor->data<float>();
if (scale_datanorm >= 0) {
if (t.find(".batch_size@GRAD") != std::string::npos ||
t.find(".batch_sum@GRAD") != std::string::npos) {
Eigen::Map<Eigen::MatrixXf> mat(g, 1, count);
float scale = 1.0 / batch_size;
mat *= scale;
} else if (t.find(".batch_square_sum@GRAD") != std::string::npos) {
VLOG(3) << "epsilon: " << scale_datanorm;
for (int i = 0; i < count; ++i) {
g[i] = (g[i] - batch_size * scale_datanorm) / batch_size +
batch_size * scale_datanorm;
}
}
}
paddle::ps::Region reg(g, count);
regions.emplace_back(std::move(reg));
}
......@@ -508,18 +523,29 @@ void FleetWrapper::ShrinkSparseTable(int table_id) {
void FleetWrapper::ShrinkDenseTable(int table_id, Scope* scope,
std::vector<std::string> var_list,
float decay) {
float decay, int emb_dim) {
#ifdef PADDLE_WITH_PSLIB
std::vector<paddle::ps::Region> regions;
for (std::string& name : var_list) {
if (name.find("batch_sum") != std::string::npos) {
Variable* var = scope->FindVar(name);
CHECK(var != nullptr) << "var[" << name << "] not found";
VLOG(3) << "prepare shrink dense batch_sum";
VLOG(0) << "prepare shrink dense batch_sum";
LoDTensor* tensor = var->GetMutable<LoDTensor>();
float* g = tensor->data<float>();
Eigen::Map<Eigen::MatrixXf> mat(g, 1, tensor->numel());
mat *= decay;
// show_batch_sum += N * log(decay)
std::string size_name = name;
size_name.replace(size_name.find("batch_sum"), size_name.length(),
"batch_size");
Variable* var_size = scope->FindVar(size_name);
CHECK(var_size != nullptr) << "var[" << size_name << "] not found";
VLOG(3) << "shrink dense batch_sum: " << name << ", " << size_name;
float* g_size = var_size->GetMutable<LoDTensor>()->data<float>();
for (int k = 0; k < tensor->numel(); k += emb_dim) {
g[k] = g[k] + g_size[k] * log(decay);
}
paddle::ps::Region reg(g, tensor->numel());
regions.emplace_back(std::move(reg));
} else {
......
......@@ -82,7 +82,8 @@ class FleetWrapper {
void PushDenseVarsAsync(
const Scope& scope, const uint64_t table_id,
const std::vector<std::string>& var_names,
std::vector<::std::future<int32_t>>* push_sparse_status);
std::vector<::std::future<int32_t>>* push_sparse_status,
float scale_datanorm, int batch_size);
void PushDenseVarsSync(Scope* scope, const uint64_t table_id,
const std::vector<std::string>& var_names);
......@@ -149,7 +150,8 @@ class FleetWrapper {
void ShrinkSparseTable(int table_id);
void ShrinkDenseTable(int table_id, Scope* scope,
std::vector<std::string> var_list, float decay);
std::vector<std::string> var_list, float decay,
int emb_dim);
// register client to client communication
typedef std::function<int32_t(int, int, const std::string&)> MsgHandlerFunc;
......
......@@ -34,6 +34,7 @@ message TrainerDesc {
optional FetchConfig fetch_config = 7;
optional bool use_cvm = 8 [ default = false ];
optional bool dump_slot = 9 [ default = false ];
optional float scale_datanorm = 10 [ default = -1 ];
// device worker parameters
optional HogwildWorkerParameter hogwild_param = 101;
......
......@@ -227,21 +227,22 @@ class PSLib(Fleet):
self._fleet_ptr.shrink_sparse_table(i.table_id)
self._role_maker._barrier_worker()
def shrink_dense_table(self, decay, scope=None, table_id=None):
def shrink_dense_table(self, decay, emb_dim=11, scope=None, table_id=None):
"""
shrink all dense params in pserver by multiplying by decay
shrink batch_sum in pserver by multiplying by decay
Args:
decay(float): the decay rate, usually range in (0, 1)
emb_dim(int): one element's length in datanorm layer
scope(Scope): Scope object, default is fluid.global_scope()
table_id(int): table id of shrinking dense table. None means shrink all,
you should specify it when using multiple scopes,
default is None.
Example:
>>> fleet.shrink_dense_table(0.98, myscope1, 1)
>>> fleet.shrink_dense_table(0.98, myscope1, 2)
>>> fleet.shrink_dense_table(0.98, myscope2, 3)
>>> fleet.shrink_dense_table(0.98, 11, myscope1, 1)
>>> fleet.shrink_dense_table(0.98, 11, myscope1, 2)
>>> fleet.shrink_dense_table(0.98, 11, myscope2, 3)
"""
if scope is None:
......@@ -260,7 +261,7 @@ class PSLib(Fleet):
if skip:
continue
self._fleet_ptr.shrink_dense_table(i.table_id, scope, var_list,
decay)
decay, emb_dim)
self._role_maker._barrier_worker()
def load_one_table(self, table_id, model_path, **kwargs):
......
......@@ -162,6 +162,7 @@ class DistributedAdam(DistributedOptimizerImplBase):
opt_info["fleet_desc"] = ps_param
opt_info["worker_skipped_ops"] = worker_skipped_ops
opt_info["use_cvm"] = strategy.get("use_cvm", False)
opt_info["scale_datanorm"] = strategy.get("scale_datanorm", -1)
opt_info["dump_slot"] = False
if server._server.downpour_server_param.downpour_table_param[
0].accessor.accessor_class == "DownpourCtrAccessor":
......
......@@ -75,6 +75,9 @@ class TrainerDesc(object):
def _set_use_cvm(self, use_cvm=False):
self.proto_desc.use_cvm = use_cvm
def _set_scale_datanorm(self, scale_datanorm=-1):
self.proto_desc.scale_datanorm = scale_datanorm
def _set_dump_slot(self, dump_slot):
self.proto_desc.dump_slot = dump_slot
......
......@@ -39,6 +39,7 @@ class TrainerFactory(object):
device_worker._set_fleet_desc(opt_info["fleet_desc"])
trainer._set_fleet_desc(opt_info["fleet_desc"])
trainer._set_use_cvm(opt_info["use_cvm"])
trainer._set_scale_datanorm(opt_info["scale_datanorm"])
trainer._set_dump_slot(opt_info["dump_slot"])
trainer._set_device_worker(device_worker)
return trainer
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册