Created by: ViktorAxelsen

• 版本、环境信息：    1）PaddlePaddle版本：1.6.2    2）CPU：AI Studio    3）GPU：V100，AI Studio上的    4）系统环境：Python2.7
• 训练信息    1）单卡    2）16GB / 32GB    3）linear_lr_warmup
• 复现信息：环境就是AI Studio，代码为AI识虫比赛官方给的train.py代码
• 问题描述：动态图下，训练时使用linear_lr_warmup报错： Traceback (most recent call last): File "train.py", line 58, in decayed_lr = fluid.layers.linear_lr_warmup(learning_rate, 850, 1e-6, 1e-3) File "/opt/conda/envs/python27-paddle120-env/lib/python2.7/site-packages/paddle/fluid/layers/learning_rate_scheduler.py", line 535, in linear_lr_warmup tensor.assign(decayed_lr, lr) File "/opt/conda/envs/python27-paddle120-env/lib/python2.7/site-packages/paddle/fluid/layers/control_flow.py", line 1595, in exit self.block.complete() File "/opt/conda/envs/python27-paddle120-env/lib/python2.7/site-packages/paddle/fluid/layers/control_flow.py", line 1680, in complete 'is_scalar_condition': self.is_scalar_condition File "/opt/conda/envs/python27-paddle120-env/lib/python2.7/site-packages/paddle/fluid/framework.py", line 2479, in append_op kwargs.get("stop_gradient", False)) File "/opt/conda/envs/python27-paddle120-env/lib/python2.7/site-packages/paddle/fluid/dygraph/tracer.py", line 47, in trace_op not stop_gradient) TypeError: trace(): incompatible function arguments. The following argument types are supported:
  1. (self: paddle.fluid.core_avx.Tracer, arg0: unicode, arg1: Dict[unicode, handle], arg2: Dict[unicode, handle], arg3: Dict[unicode, Variant], arg4: paddle::platform::CUDAPlace, arg5: bool) -> None
  2. (self: paddle.fluid.core_avx.Tracer, arg0: unicode, arg1: Dict[unicode, handle], arg2: Dict[unicode, handle], arg3: Dict[unicode, Variant], arg4: paddle::platform::CPUPlace, arg5: bool) -> None

Invoked with: <paddle.fluid.dygraph.tracer.Tracer object at 0x7f6cc2abaa70>, 'conditional_block', {'Input': [], 'Cond': [name tmp_2, dtype: VarType.FP32 shape: [1L] lod: {} dim: 1 layout: NCHW dtype: bool data: [1] ]}, {'Scope': [name _generated_var_0, shape: [0L], not inited], 'Out': []}, {'sub_block': idx: 1 parent_idx: 0 , 'is_scalar_condition': True}, <paddle.fluid.core_avx.CUDAPlace object at 0x7f6cc2b0d7d8>, True

部分代码： learning_rate = fluid.layers.piecewise_decay(boundaries=[13600, 20400], values=[1e-3, 1e-4, 1e-5]) decayed_lr = fluid.layers.linear_lr_warmup(learning_rate, 850, 1e-6, 1e-3) opt = fluid.optimizer.Adam(learning_rate=decayed_lr, regularization=fluid.regularizer.L2Decay(regularization_coeff=5e-4))