Fix adam mv initilizer (#2082)

* zero constant initilzer for adam m and v

* make of_format

* init adam m v beta1_t and beta2_t

* use value instead of initializer

* const float& -> const float

* update

oneflow/core/job_completer/adam_optm.cpp

@@ -5,10 +5,13 @@ namespace oneflow {
 namespace {
 
 OperatorConf GenerateAdamHelperVariableOpConf(const VariableOp& op, const std::string& name,
-                                              JobBuilder* job_builder) {
+                                              const float initial_value, JobBuilder* job_builder) {
   OperatorConf helper_variable_op(op.op_conf());
   helper_variable_op.set_name(op.op_name() + "-" + name);
   helper_variable_op.mutable_variable_conf()->set_out("out");
+  InitializerConf constant_initializer;
+  constant_initializer.mutable_constant_conf()->set_value(initial_value);
+  *(helper_variable_op.mutable_variable_conf()->mutable_initializer()) = constant_initializer;
   BindTwoVariableOpObnSbpConf(helper_variable_op.name(), op.op_name(), job_builder);
   return helper_variable_op;
 }
@@ -24,8 +27,8 @@ void SetScalarShapeAndSbpConf(OperatorConf* op_conf, JobBuilder* job_builder) {
 void GenerateOptimizerOpConf(const VariableOp& op, const ParallelConf& parallel_conf,
                              JobBuilder* job_builder, const LogicalBlobId& diff_lbi_of_var_out,
                              const LogicalBlobId& total_loss_instance_num_lbi) {
-  const OperatorConf& m_var = GenerateAdamHelperVariableOpConf(op, "m", job_builder);
-  const OperatorConf& v_var = GenerateAdamHelperVariableOpConf(op, "v", job_builder);
+  const OperatorConf& m_var = GenerateAdamHelperVariableOpConf(op, "m", 0.f, job_builder);
+  const OperatorConf& v_var = GenerateAdamHelperVariableOpConf(op, "v", 0.f, job_builder);
   job_builder->AddOps(parallel_conf, {m_var, v_var});
 
   OperatorConf mdupdt_op;
@@ -35,9 +38,10 @@ void GenerateOptimizerOpConf(const VariableOp& op, const ParallelConf& parallel_
       GlobalJobDesc().job_conf().train_conf().model_update_conf();
   OperatorConf beta1_t_var;
   OperatorConf beta2_t_var;
-  if (mdupdt_op_conf->user_conf().adam_conf().do_bias_correction()) {
-    beta1_t_var = GenerateAdamHelperVariableOpConf(op, "beta1_t", job_builder);
-    beta2_t_var = GenerateAdamHelperVariableOpConf(op, "beta2_t", job_builder);
+  const AdamModelUpdateConf& adam_conf = mdupdt_op_conf->user_conf().adam_conf();
+  if (adam_conf.do_bias_correction()) {
+    beta1_t_var = GenerateAdamHelperVariableOpConf(op, "beta1_t", adam_conf.beta1(), job_builder);
+    beta2_t_var = GenerateAdamHelperVariableOpConf(op, "beta2_t", adam_conf.beta2(), job_builder);
     job_builder->AddOps(parallel_conf, {beta1_t_var, beta2_t_var});
     SetScalarShapeAndSbpConf(&beta1_t_var, job_builder);
     SetScalarShapeAndSbpConf(&beta2_t_var, job_builder);
@@ -45,7 +49,7 @@ void GenerateOptimizerOpConf(const VariableOp& op, const ParallelConf& parallel_
   ConstructMdUpdtOpConf(op, diff_lbi_of_var_out, total_loss_instance_num_lbi, mdupdt_op_conf);
   mdupdt_op_conf->set_m(m_var.name() + "/out");
   mdupdt_op_conf->set_v(v_var.name() + "/out");
-  if (mdupdt_op_conf->user_conf().adam_conf().do_bias_correction()) {
+  if (adam_conf.do_bias_correction()) {
     mdupdt_op_conf->set_beta1_t(beta1_t_var.name() + "/out");
     mdupdt_op_conf->set_beta2_t(beta2_t_var.name() + "/out");
   }