未验证 提交 445d7337 编写于 作者: H hong 提交者: GitHub

fix new ir optimizer bug (#55910)

上级 37c487e0
......@@ -9,7 +9,7 @@
kernel :
func : adadelta
data_type : param
optional : master_param
optional : master_param, master_param_out
inplace : (param -> param_out), (avg_squared_grad -> moment_out), (avg_squared_update -> inf_norm_out), (master_param -> master_param_out)
- op : add
......
......@@ -44,13 +44,19 @@
extra :
attrs : [bool use_mkldnn = false, bool use_cudnn = false]
- op : adagrad_
- op : adadelta_ (adadelta)
inputs :
{param : Param, grad: Grad, avg_squared_grad : AvgSquaredGrad, avg_squared_update : AvgSquaredUpdate, learning_rate : LearningRate, master_param : MasterParam }
outputs :
{param_out : ParamOut, moment_out : AvgSquaredGradOut, inf_norm_out : AvgSquaredUpdateOut, master_param_out : MasterParamOut}
- op : adagrad_ (adagrad)
inputs :
{ param : Param, grad : Grad, moment : Moment, learning_rate : LearningRate, master_param : MasterParam }
outputs :
{ param_out : ParamOut, moment_out : MomentOut, master_param_out : MasterParamOut }
- op : adam_
- op : adam_ (adam)
inputs :
{param: Param, grad: Grad, learning_rate: LearningRate, moment1: Moment1, moment2: Moment2, beta1_pow: Beta1Pow, beta2_pow: Beta2Pow, master_param: MasterParam, skip_update: SkipUpdate}
outputs :
......@@ -67,13 +73,13 @@
tensor_name : EpsilonTensor
manual_signature : [adam_]
- op : adamax_
- op : adamax_ (adamax)
inputs :
{param : Param, grad: Grad, learning_rate : LearningRate, moment : Moment, inf_norm : InfNorm, beta1_pow : Beta1Pow, master_param : MasterParam}
outputs :
{param_out : ParamOut, moment_out : MomentOut, inf_norm_out : InfNormOut, master_param_out : MasterParamOut}
- op : adamw_
- op : adamw_ (adamw)
inputs :
{param: Param, grad: Grad, learning_rate: LearningRate, moment1: Moment1, moment2: Moment2, beta1_pow: Beta1Pow, beta2_pow: Beta2Pow, master_param: MasterParam, skip_update: SkipUpdate}
outputs :
......@@ -1454,7 +1460,7 @@
outputs :
out : Out
- op : lamb_
- op : lamb_ (lamb)
inputs :
{param : Param, grad : Grad, learning_rate : LearningRate, moment1 : Moment1, moment2 : Moment2, beta1_pow : Beta1Pow, beta2_pow : Beta2Pow, master_param : MasterParam, skip_update : SkipUpdate}
outputs :
......@@ -1859,7 +1865,7 @@
outputs :
{out : Out, indices : Indices}
- op : momentum_
- op : momentum_ (momentum)
inputs :
{param : Param, grad : Grad, velocity : Velocity, learning_rate : LearningRate, master_param : MasterParam}
outputs :
......@@ -2264,7 +2270,7 @@
support_tensor : true
manual_signature : [reverse]
- op : rmsprop_
- op : rmsprop_ (rmsprop)
inputs :
{param: Param, mean_square: MeanSquare, mean_grad: MeanGrad, learning_rate: LearningRate, grad: Grad, moment: Moment, master_param: MasterParam}
outputs :
......@@ -2402,7 +2408,7 @@
extra :
attrs : [str data_format = "AnyLayout"]
- op : sgd_
- op : sgd_ (sgd)
inputs :
{param : Param, learning_rate : LearningRate, grad : Grad, master_param : MasterParam}
outputs :
......
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