Created by: Xreki

If user uses a op which does not have grad_op, and forgets to set stop_gradient to True, the train process will crash.

The error message is:

Traceback (most recent call last):
  File "./train_gpu_paddle.py", line 471, in <module>
    train(n_token, cutoffs)
  File "./train_gpu_paddle.py", line 254, in train
    decr_ratio=FLAGS.decr_ratio)
  File "/home/dingsiyu/project/python/transformer-xl-paddlepaddle/optimization.py", line 151, in optimization
    _, param_grads = optimizer.minimize(loss)
  File "<decorator-gen-20>", line 2, in minimize
  File "/home/dingsiyu/bin/anaconda3/lib/python3.6/site-packages/paddle/fluid/wrapped_decorator.py", line 25, in __impl__
    return wrapped_func(*args, **kwargs)
  File "/home/dingsiyu/bin/anaconda3/lib/python3.6/site-packages/paddle/fluid/dygraph/base.py", line 88, in __impl__
    return func(*args, **kwargs)
  File "/home/dingsiyu/bin/anaconda3/lib/python3.6/site-packages/paddle/fluid/optimizer.py", line 593, in minimize
    no_grad_set=no_grad_set)
  File "/home/dingsiyu/bin/anaconda3/lib/python3.6/site-packages/paddle/fluid/optimizer.py", line 493, in backward
    no_grad_set, callbacks)
  File "/home/dingsiyu/bin/anaconda3/lib/python3.6/site-packages/paddle/fluid/backward.py", line 570, in append_backward
    input_grad_names_set=input_grad_names_set)
  File "/home/dingsiyu/bin/anaconda3/lib/python3.6/site-packages/paddle/fluid/backward.py", line 310, in _append_backward_ops_
    op.desc, cpt.to_text(no_grad_dict[block.idx]), grad_sub_block_list)
paddle.fluid.core_avx.EnforceNotMet: grad_op_maker_ should not be null
Operator GradOpMaker has not been registered. at [/paddle/paddle/fluid/framework/op_info.h:69]
PaddlePaddle Call Stacks: 
0       0x7ff346440948p void paddle::platform::EnforceNotMet::Init<std::string>(std::string, char const*, int) + 360
1       0x7ff346440c97p paddle::platform::EnforceNotMet::EnforceNotMet(std::string const&, char const*, int) + 87
2       0x7ff346441c5cp paddle::framework::OpInfo::GradOpMaker() const + 108
3       0x7ff34643935ep
4       0x7ff346472de6p
5       0x7ff381de7744p _PyCFunction_FastCallDict + 340
6       0x7ff381e7593cp
7       0x7ff381e99a7ap _PyEval_EvalFrameDefault + 762

There is no useful information for users to fix. After this PR, error message is:

Traceback (most recent call last):
  File "./train_gpu_paddle.py", line 471, in <module>
    train(n_token, cutoffs)
  File "./train_gpu_paddle.py", line 254, in train
    decr_ratio=FLAGS.decr_ratio)
  File "/work/Paddle/build_paddle/test/users/transformer-xl-paddlepaddle-lyq/optimization.py", line 151, in optimization
    _, param_grads = optimizer.minimize(loss)
  File "</usr/local/lib/python2.7/dist-packages/decorator.pyc:decorator-gen-20>", line 2, in minimize
  File "/usr/local/lib/python2.7/dist-packages/paddle/fluid/wrapped_decorator.py", line 25, in __impl__
    return wrapped_func(*args, **kwargs)
  File "/usr/local/lib/python2.7/dist-packages/paddle/fluid/dygraph/base.py", line 86, in __impl__
    return func(*args, **kwargs)
  File "/usr/local/lib/python2.7/dist-packages/paddle/fluid/optimizer.py", line 594, in minimize
    no_grad_set=no_grad_set)
  File "/usr/local/lib/python2.7/dist-packages/paddle/fluid/optimizer.py", line 493, in backward
    no_grad_set, callbacks)
  File "/usr/local/lib/python2.7/dist-packages/paddle/fluid/backward.py", line 699, in append_backward
    input_grad_names_set=input_grad_names_set)
  File "/usr/local/lib/python2.7/dist-packages/paddle/fluid/backward.py", line 432, in _append_backward_ops_
    op.desc, cpt.to_text(no_grad_dict[block.idx]), grad_sub_block_list)
paddle.fluid.core_avx.EnforceNotMet: grad_op_maker_ should not be null 
Operator range's GradOpMaker has not been registered.                                                                                                                                        
Please check whether range_op has grad_op.
If not, please set stop_gradient to True for its input and output variables using var.step_gradient=True. at [/paddle/paddle/fluid/framework/op_info.h:75]
PaddlePaddle Call Stacks: 
0       0x7f6ffa355188p void paddle::platform::EnforceNotMet::Init<std::string>(std::string, char const*, int) + 360
1       0x7f6ffa3554d7p paddle::platform::EnforceNotMet::EnforceNotMet(std::string const&, char const*, int) + 87 
2       0x7f6ffa356ad9p paddle::framework::OpInfo::GradOpMaker() const + 233
3       0x7f6ffa34db9ep
4       0x7f6ffa3881e6p