Using cuDNN version 5110 on context None
Mapped name None to device cuda1: TITAN X (Pascal) (0000:04:00.0)
ERROR (theano.gof.opt): Optimization failure due to: local_gpu_elemwise_careduce
ERROR (theano.gof.opt): node: GpuCAReduceCuda{add}{0, 2, 3}(GpuElemwise{sqr,no_inplace}.0)
ERROR (theano.gof.opt): TRACEBACK:
ERROR (theano.gof.opt): Traceback (most recent call last):
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/opt.py", line 2036, in process_node
    remove=remove)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 569, in replace_all_validate_remove
    chk = fgraph.replace_all_validate(replacements, reason)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 518, in replace_all_validate
    fgraph.replace(r, new_r, reason=reason, verbose=False)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/fg.py", line 486, in replace
    ". The type of the replacement must be the same.", old, new)
BadOptimization: BadOptimization Error 
  Variable: id 140341636258832 GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}.0
  Op GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}(GpuElemwise{sub,no_inplace}.0)
  Value Type: <type 'NoneType'>
  Old Value:  None
  New Value:  None
  Reason:  local_gpu_elemwise_careduce. The type of the replacement must be the same.
  Old Graph:
  GpuCAReduceCuda{add}{0, 2, 3} [id A] <GpuArrayType<None>(float32, vector)> ''   
   |GpuElemwise{sqr,no_inplace} [id B] <GpuArrayType<None>(float16, 4D)> ''   
     |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   
       |AbstractConv2d{convdim=2, border_mode='half', subsample=(2, 2), filter_flip=True, imshp=(None, 128, 8, 8), kshp=(128, 128, 3, 3), filter_dilation=(1, 1)} [id D] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuElemwise{add,no_inplace} [id E] <GpuArrayType<None>(float16, 4D)> ''   
       | | |GpuElemwise{mul,no_inplace} [id F] <GpuArrayType<None>(float16, 4D)> ''   
       | | |GpuElemwise{mul,no_inplace} [id G] <GpuArrayType<None>(float16, 4D)> ''   
       | |W [id H] <GpuArrayType<None>(float16, 4D)>
       |GpuElemwise{Cast{float16}}[]<gpuarray> [id I] <GpuArrayType<None>(float16, (True, False, True, True))> ''   
         |GpuElemwise{true_div,no_inplace} [id J] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuElemwise{true_div,no_inplace} [id K] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuFromHost<None> [id L] <GpuArrayType<None>(float32, (True, True, True, True))> ''   

  New Graph:
  GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3} [id M] <GpuArrayType<None>(float16, vector)> ''   
   |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   


Hint: relax the tolerance by setting tensor.cmp_sloppy=1
  or even tensor.cmp_sloppy=2 for less-strict comparison


ERROR (theano.gof.opt): Optimization failure due to: local_gpu_elemwise_careduce
ERROR (theano.gof.opt): node: GpuCAReduceCuda{add}{0, 2, 3}(GpuElemwise{sqr,no_inplace}.0)
ERROR (theano.gof.opt): TRACEBACK:
ERROR (theano.gof.opt): Traceback (most recent call last):
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/opt.py", line 2036, in process_node
    remove=remove)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 569, in replace_all_validate_remove
    chk = fgraph.replace_all_validate(replacements, reason)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 518, in replace_all_validate
    fgraph.replace(r, new_r, reason=reason, verbose=False)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/fg.py", line 486, in replace
    ". The type of the replacement must be the same.", old, new)
BadOptimization: BadOptimization Error 
  Variable: id 140341635638928 GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}.0
  Op GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}(GpuElemwise{sub,no_inplace}.0)
  Value Type: <type 'NoneType'>
  Old Value:  None
  New Value:  None
  Reason:  local_gpu_elemwise_careduce. The type of the replacement must be the same.
  Old Graph:
  GpuCAReduceCuda{add}{0, 2, 3} [id A] <GpuArrayType<None>(float32, vector)> ''   
   |GpuElemwise{sqr,no_inplace} [id B] <GpuArrayType<None>(float16, 4D)> ''   
     |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   
       |AbstractConv2d{convdim=2, border_mode='half', subsample=(1, 1), filter_flip=True, imshp=(None, 64, 8, 8), kshp=(128, 64, 3, 3), filter_dilation=(1, 1)} [id D] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuElemwise{add,no_inplace} [id E] <GpuArrayType<None>(float16, 4D)> ''   
       | | |GpuElemwise{mul,no_inplace} [id F] <GpuArrayType<None>(float16, 4D)> ''   
       | | |GpuElemwise{mul,no_inplace} [id G] <GpuArrayType<None>(float16, 4D)> ''   
       | |W [id H] <GpuArrayType<None>(float16, 4D)>
       |GpuElemwise{Cast{float16}}[]<gpuarray> [id I] <GpuArrayType<None>(float16, (True, False, True, True))> ''   
         |GpuElemwise{true_div,no_inplace} [id J] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuElemwise{true_div,no_inplace} [id K] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuFromHost<None> [id L] <GpuArrayType<None>(float32, (True, True, True, True))> ''   

  New Graph:
  GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3} [id M] <GpuArrayType<None>(float16, vector)> ''   
   |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   


Hint: relax the tolerance by setting tensor.cmp_sloppy=1
  or even tensor.cmp_sloppy=2 for less-strict comparison


ERROR (theano.gof.opt): Optimization failure due to: local_gpu_elemwise_careduce
ERROR (theano.gof.opt): node: GpuCAReduceCuda{add}{0, 2, 3}(GpuElemwise{sqr,no_inplace}.0)
ERROR (theano.gof.opt): TRACEBACK:
ERROR (theano.gof.opt): Traceback (most recent call last):
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/opt.py", line 2036, in process_node
    remove=remove)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 569, in replace_all_validate_remove
    chk = fgraph.replace_all_validate(replacements, reason)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 518, in replace_all_validate
    fgraph.replace(r, new_r, reason=reason, verbose=False)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/fg.py", line 486, in replace
    ". The type of the replacement must be the same.", old, new)
BadOptimization: BadOptimization Error 
  Variable: id 140341635312912 GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}.0
  Op GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}(GpuElemwise{sub,no_inplace}.0)
  Value Type: <type 'NoneType'>
  Old Value:  None
  New Value:  None
  Reason:  local_gpu_elemwise_careduce. The type of the replacement must be the same.
  Old Graph:
  GpuCAReduceCuda{add}{0, 2, 3} [id A] <GpuArrayType<None>(float32, vector)> ''   
   |GpuElemwise{sqr,no_inplace} [id B] <GpuArrayType<None>(float16, 4D)> ''   
     |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   
       |AbstractConv2d{convdim=2, border_mode='half', subsample=(2, 2), filter_flip=True, imshp=(None, 64, 16, 16), kshp=(64, 64, 3, 3), filter_dilation=(1, 1)} [id D] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuElemwise{add,no_inplace} [id E] <GpuArrayType<None>(float16, 4D)> ''   
       | | |GpuElemwise{mul,no_inplace} [id F] <GpuArrayType<None>(float16, 4D)> ''   
       | | |GpuElemwise{mul,no_inplace} [id G] <GpuArrayType<None>(float16, 4D)> ''   
       | |W [id H] <GpuArrayType<None>(float16, 4D)>
       |GpuElemwise{Cast{float16}}[]<gpuarray> [id I] <GpuArrayType<None>(float16, (True, False, True, True))> ''   
         |GpuElemwise{true_div,no_inplace} [id J] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuElemwise{true_div,no_inplace} [id K] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuFromHost<None> [id L] <GpuArrayType<None>(float32, (True, True, True, True))> ''   

  New Graph:
  GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3} [id M] <GpuArrayType<None>(float16, vector)> ''   
   |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   


Hint: relax the tolerance by setting tensor.cmp_sloppy=1
  or even tensor.cmp_sloppy=2 for less-strict comparison


ERROR (theano.gof.opt): Optimization failure due to: local_gpu_elemwise_careduce
ERROR (theano.gof.opt): node: GpuCAReduceCuda{add}{0, 2, 3}(GpuElemwise{sqr,no_inplace}.0)
ERROR (theano.gof.opt): TRACEBACK:
ERROR (theano.gof.opt): Traceback (most recent call last):
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/opt.py", line 2036, in process_node
    remove=remove)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 569, in replace_all_validate_remove
    chk = fgraph.replace_all_validate(replacements, reason)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 518, in replace_all_validate
    fgraph.replace(r, new_r, reason=reason, verbose=False)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/fg.py", line 486, in replace
    ". The type of the replacement must be the same.", old, new)
BadOptimization: BadOptimization Error 
  Variable: id 140341634976336 GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}.0
  Op GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}(GpuElemwise{sub,no_inplace}.0)
  Value Type: <type 'NoneType'>
  Old Value:  None
  New Value:  None
  Reason:  local_gpu_elemwise_careduce. The type of the replacement must be the same.
  Old Graph:
  GpuCAReduceCuda{add}{0, 2, 3} [id A] <GpuArrayType<None>(float32, vector)> ''   
   |GpuElemwise{sqr,no_inplace} [id B] <GpuArrayType<None>(float16, 4D)> ''   
     |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   
       |AbstractConv2d{convdim=2, border_mode='half', subsample=(1, 1), filter_flip=True, imshp=(None, 32, 16, 16), kshp=(64, 32, 3, 3), filter_dilation=(1, 1)} [id D] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuElemwise{add,no_inplace} [id E] <GpuArrayType<None>(float16, 4D)> ''   
       | | |GpuElemwise{mul,no_inplace} [id F] <GpuArrayType<None>(float16, 4D)> ''   
       | | |GpuElemwise{mul,no_inplace} [id G] <GpuArrayType<None>(float16, 4D)> ''   
       | |W [id H] <GpuArrayType<None>(float16, 4D)>
       |GpuElemwise{Cast{float16}}[]<gpuarray> [id I] <GpuArrayType<None>(float16, (True, False, True, True))> ''   
         |GpuElemwise{true_div,no_inplace} [id J] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuElemwise{true_div,no_inplace} [id K] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuFromHost<None> [id L] <GpuArrayType<None>(float32, (True, True, True, True))> ''   

  New Graph:
  GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3} [id M] <GpuArrayType<None>(float16, vector)> ''   
   |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   


Hint: relax the tolerance by setting tensor.cmp_sloppy=1
  or even tensor.cmp_sloppy=2 for less-strict comparison


ERROR (theano.gof.opt): Optimization failure due to: local_gpu_elemwise_careduce
ERROR (theano.gof.opt): node: GpuCAReduceCuda{add}{0, 2, 3}(GpuElemwise{sqr,no_inplace}.0)
ERROR (theano.gof.opt): TRACEBACK:
ERROR (theano.gof.opt): Traceback (most recent call last):
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/opt.py", line 2036, in process_node
    remove=remove)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 569, in replace_all_validate_remove
    chk = fgraph.replace_all_validate(replacements, reason)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 518, in replace_all_validate
    fgraph.replace(r, new_r, reason=reason, verbose=False)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/fg.py", line 486, in replace
    ". The type of the replacement must be the same.", old, new)
BadOptimization: BadOptimization Error 
  Variable: id 140341635160400 GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}.0
  Op GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}(GpuElemwise{sub,no_inplace}.0)
  Value Type: <type 'NoneType'>
  Old Value:  None
  New Value:  None
  Reason:  local_gpu_elemwise_careduce. The type of the replacement must be the same.
  Old Graph:
  GpuCAReduceCuda{add}{0, 2, 3} [id A] <GpuArrayType<None>(float32, vector)> ''   
   |GpuElemwise{sqr,no_inplace} [id B] <GpuArrayType<None>(float16, 4D)> ''   
     |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   
       |AbstractConv2d{convdim=2, border_mode='half', subsample=(2, 2), filter_flip=True, imshp=(None, 32, 32, 32), kshp=(32, 32, 3, 3), filter_dilation=(1, 1)} [id D] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuElemwise{add,no_inplace} [id E] <GpuArrayType<None>(float16, 4D)> ''   
       | | |GpuElemwise{mul,no_inplace} [id F] <GpuArrayType<None>(float16, 4D)> ''   
       | | |GpuElemwise{mul,no_inplace} [id G] <GpuArrayType<None>(float16, 4D)> ''   
       | |W [id H] <GpuArrayType<None>(float16, 4D)>
       |GpuElemwise{Cast{float16}}[]<gpuarray> [id I] <GpuArrayType<None>(float16, (True, False, True, True))> ''   
         |GpuElemwise{true_div,no_inplace} [id J] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuElemwise{true_div,no_inplace} [id K] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuFromHost<None> [id L] <GpuArrayType<None>(float32, (True, True, True, True))> ''   

  New Graph:
  GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3} [id M] <GpuArrayType<None>(float16, vector)> ''   
   |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   


Hint: relax the tolerance by setting tensor.cmp_sloppy=1
  or even tensor.cmp_sloppy=2 for less-strict comparison


ERROR (theano.gof.opt): Optimization failure due to: local_gpu_elemwise_careduce
ERROR (theano.gof.opt): node: GpuCAReduceCuda{add}{0, 2, 3}(GpuElemwise{sqr,no_inplace}.0)
ERROR (theano.gof.opt): TRACEBACK:
ERROR (theano.gof.opt): Traceback (most recent call last):
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/opt.py", line 2036, in process_node
    remove=remove)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 569, in replace_all_validate_remove
    chk = fgraph.replace_all_validate(replacements, reason)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 518, in replace_all_validate
    fgraph.replace(r, new_r, reason=reason, verbose=False)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/fg.py", line 486, in replace
    ". The type of the replacement must be the same.", old, new)
BadOptimization: BadOptimization Error 
  Variable: id 140341634738000 GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}.0
  Op GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}(GpuElemwise{sub,no_inplace}.0)
  Value Type: <type 'NoneType'>
  Old Value:  None
  New Value:  None
  Reason:  local_gpu_elemwise_careduce. The type of the replacement must be the same.
  Old Graph:
  GpuCAReduceCuda{add}{0, 2, 3} [id A] <GpuArrayType<None>(float32, vector)> ''   
   |GpuElemwise{sqr,no_inplace} [id B] <GpuArrayType<None>(float16, 4D)> ''   
     |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   
       |AbstractConv2d{convdim=2, border_mode='half', subsample=(1, 1), filter_flip=True, imshp=(None, 3, 32, 32), kshp=(32, 3, 3, 3), filter_dilation=(1, 1)} [id D] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuFromHost<None> [id E] <GpuArrayType<None>(float16, 4D)> ''   
       | | |<TensorType(float16, 4D)> [id F] <TensorType(float16, 4D)>
       | |W [id G] <GpuArrayType<None>(float16, 4D)>
       |GpuElemwise{Cast{float16}}[]<gpuarray> [id H] <GpuArrayType<None>(float16, (True, False, True, True))> ''   
         |GpuElemwise{true_div,no_inplace} [id I] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuElemwise{true_div,no_inplace} [id J] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuFromHost<None> [id K] <GpuArrayType<None>(float32, (True, True, True, True))> ''   

  New Graph:
  GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3} [id L] <GpuArrayType<None>(float16, vector)> ''   
   |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   


Hint: relax the tolerance by setting tensor.cmp_sloppy=1
  or even tensor.cmp_sloppy=2 for less-strict comparison


ERROR (theano.gof.opt): Optimization failure due to: local_gpu_elemwise_careduce
ERROR (theano.gof.opt): node: GpuCAReduceCuda{add}{0, 2, 3}(GpuElemwise{sqr,no_inplace}.0)
ERROR (theano.gof.opt): TRACEBACK:
ERROR (theano.gof.opt): Traceback (most recent call last):
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/opt.py", line 2036, in process_node
    remove=remove)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 569, in replace_all_validate_remove
    chk = fgraph.replace_all_validate(replacements, reason)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 518, in replace_all_validate
    fgraph.replace(r, new_r, reason=reason, verbose=False)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/fg.py", line 486, in replace
    ". The type of the replacement must be the same.", old, new)
BadOptimization: BadOptimization Error 
  Variable: id 140341665146704 GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}.0
  Op GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}(GpuElemwise{sub,no_inplace}.0)
  Value Type: <type 'NoneType'>
  Old Value:  None
  New Value:  None
  Reason:  local_gpu_elemwise_careduce. The type of the replacement must be the same.
  Old Graph:
  GpuCAReduceCuda{add}{0, 2, 3} [id A] <GpuArrayType<None>(float32, vector)> ''   
   |GpuElemwise{sqr,no_inplace} [id B] <GpuArrayType<None>(float16, 4D)> ''   
     |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   
       |GpuDnnConv{algo='small', inplace=False} [id D] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuContiguous [id E] <GpuArrayType<None>(float16, 4D)> ''   
       | | |GpuElemwise{add,no_inplace} [id F] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuContiguous [id G] <GpuArrayType<None>(float16, 4D)> ''   
       | | |W [id H] <GpuArrayType<None>(float16, 4D)>
       | |GpuAllocEmpty{dtype='float16', context_name=None} [id I] <GpuArrayType<None>(float16, 4D)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[0] < 0).'} [id J] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[1] < 0).'} [id K] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[2] <= 0).'} [id L] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[3] <= 0).'} [id M] <TensorType(int64, scalar)> ''   
       | |GpuDnnConvDesc{border_mode='half', subsample=(2, 2), dilation=(1, 1), conv_mode='conv', precision='float32'} [id N] <CDataType{cudnnConvolutionDescriptor_t}> ''   
       | | |Shape [id O] <TensorType(int64, vector)> ''   
       | |Cast{float32} [id P] <float32> ''   
       | | |Constant{1.0} [id Q] <float64>
       | |Cast{float32} [id R] <float32> ''   
       |   |Constant{0.0} [id S] <float64>
       |GpuElemwise{Cast{float16}}[]<gpuarray> [id T] <GpuArrayType<None>(float16, (True, False, True, True))> ''   
         |GpuElemwise{true_div,no_inplace} [id U] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuElemwise{true_div,no_inplace} [id V] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuFromHost<None> [id W] <GpuArrayType<None>(float32, (True, True, True, True))> ''   

  New Graph:
  GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3} [id X] <GpuArrayType<None>(float16, vector)> ''   
   |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   


Hint: relax the tolerance by setting tensor.cmp_sloppy=1
  or even tensor.cmp_sloppy=2 for less-strict comparison


ERROR (theano.gof.opt): Optimization failure due to: local_gpu_elemwise_careduce
ERROR (theano.gof.opt): node: GpuCAReduceCuda{add}{0, 2, 3}(GpuElemwise{sqr,no_inplace}.0)
ERROR (theano.gof.opt): TRACEBACK:
ERROR (theano.gof.opt): Traceback (most recent call last):
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/opt.py", line 2036, in process_node
    remove=remove)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 569, in replace_all_validate_remove
    chk = fgraph.replace_all_validate(replacements, reason)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 518, in replace_all_validate
    fgraph.replace(r, new_r, reason=reason, verbose=False)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/fg.py", line 486, in replace
    ". The type of the replacement must be the same.", old, new)
BadOptimization: BadOptimization Error 
  Variable: id 140341664725136 GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}.0
  Op GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}(GpuElemwise{sub,no_inplace}.0)
  Value Type: <type 'NoneType'>
  Old Value:  None
  New Value:  None
  Reason:  local_gpu_elemwise_careduce. The type of the replacement must be the same.
  Old Graph:
  GpuCAReduceCuda{add}{0, 2, 3} [id A] <GpuArrayType<None>(float32, vector)> ''   
   |GpuElemwise{sqr,no_inplace} [id B] <GpuArrayType<None>(float16, 4D)> ''   
     |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   
       |GpuDnnConv{algo='small', inplace=False} [id D] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuContiguous [id E] <GpuArrayType<None>(float16, 4D)> ''   
       | | |GpuElemwise{add,no_inplace} [id F] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuContiguous [id G] <GpuArrayType<None>(float16, 4D)> ''   
       | | |W [id H] <GpuArrayType<None>(float16, 4D)>
       | |GpuAllocEmpty{dtype='float16', context_name=None} [id I] <GpuArrayType<None>(float16, 4D)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[0] < 0).'} [id J] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[1] < 0).'} [id K] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[2] <= 0).'} [id L] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[3] <= 0).'} [id M] <TensorType(int64, scalar)> ''   
       | |GpuDnnConvDesc{border_mode='half', subsample=(1, 1), dilation=(1, 1), conv_mode='conv', precision='float32'} [id N] <CDataType{cudnnConvolutionDescriptor_t}> ''   
       | | |Shape [id O] <TensorType(int64, vector)> ''   
       | |Cast{float32} [id P] <float32> ''   
       | | |Constant{1.0} [id Q] <float64>
       | |Cast{float32} [id R] <float32> ''   
       |   |Constant{0.0} [id S] <float64>
       |GpuElemwise{Cast{float16}}[]<gpuarray> [id T] <GpuArrayType<None>(float16, (True, False, True, True))> ''   
         |GpuElemwise{true_div,no_inplace} [id U] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuElemwise{true_div,no_inplace} [id V] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuFromHost<None> [id W] <GpuArrayType<None>(float32, (True, True, True, True))> ''   

  New Graph:
  GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3} [id X] <GpuArrayType<None>(float16, vector)> ''   
   |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   


Hint: relax the tolerance by setting tensor.cmp_sloppy=1
  or even tensor.cmp_sloppy=2 for less-strict comparison


ERROR (theano.gof.opt): Optimization failure due to: local_gpu_elemwise_careduce
ERROR (theano.gof.opt): node: GpuCAReduceCuda{add}{0, 2, 3}(GpuElemwise{sqr,no_inplace}.0)
ERROR (theano.gof.opt): TRACEBACK:
ERROR (theano.gof.opt): Traceback (most recent call last):
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/opt.py", line 2036, in process_node
    remove=remove)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 569, in replace_all_validate_remove
    chk = fgraph.replace_all_validate(replacements, reason)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 518, in replace_all_validate
    fgraph.replace(r, new_r, reason=reason, verbose=False)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/fg.py", line 486, in replace
    ". The type of the replacement must be the same.", old, new)
BadOptimization: BadOptimization Error 
  Variable: id 140341631691408 GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}.0
  Op GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}(GpuElemwise{sub,no_inplace}.0)
  Value Type: <type 'NoneType'>
  Old Value:  None
  New Value:  None
  Reason:  local_gpu_elemwise_careduce. The type of the replacement must be the same.
  Old Graph:
  GpuCAReduceCuda{add}{0, 2, 3} [id A] <GpuArrayType<None>(float32, vector)> ''   
   |GpuElemwise{sqr,no_inplace} [id B] <GpuArrayType<None>(float16, 4D)> ''   
     |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   
       |GpuDnnConv{algo='small', inplace=False} [id D] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuContiguous [id E] <GpuArrayType<None>(float16, 4D)> ''   
       | | |GpuElemwise{add,no_inplace} [id F] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuContiguous [id G] <GpuArrayType<None>(float16, 4D)> ''   
       | | |W [id H] <GpuArrayType<None>(float16, 4D)>
       | |GpuAllocEmpty{dtype='float16', context_name=None} [id I] <GpuArrayType<None>(float16, 4D)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[0] < 0).'} [id J] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[1] < 0).'} [id K] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[2] <= 0).'} [id L] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[3] <= 0).'} [id M] <TensorType(int64, scalar)> ''   
       | |GpuDnnConvDesc{border_mode='half', subsample=(2, 2), dilation=(1, 1), conv_mode='conv', precision='float32'} [id N] <CDataType{cudnnConvolutionDescriptor_t}> ''   
       | | |Shape [id O] <TensorType(int64, vector)> ''   
       | |Cast{float32} [id P] <float32> ''   
       | | |Constant{1.0} [id Q] <float64>
       | |Cast{float32} [id R] <float32> ''   
       |   |Constant{0.0} [id S] <float64>
       |GpuElemwise{Cast{float16}}[]<gpuarray> [id T] <GpuArrayType<None>(float16, (True, False, True, True))> ''   
         |GpuElemwise{true_div,no_inplace} [id U] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuElemwise{true_div,no_inplace} [id V] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuFromHost<None> [id W] <GpuArrayType<None>(float32, (True, True, True, True))> ''   

  New Graph:
  GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3} [id X] <GpuArrayType<None>(float16, vector)> ''   
   |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   


Hint: relax the tolerance by setting tensor.cmp_sloppy=1
  or even tensor.cmp_sloppy=2 for less-strict comparison


ERROR (theano.gof.opt): Optimization failure due to: local_gpu_elemwise_careduce
ERROR (theano.gof.opt): node: GpuCAReduceCuda{add}{0, 2, 3}(GpuElemwise{sqr,no_inplace}.0)
ERROR (theano.gof.opt): TRACEBACK:
ERROR (theano.gof.opt): Traceback (most recent call last):
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/opt.py", line 2036, in process_node
    remove=remove)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 569, in replace_all_validate_remove
    chk = fgraph.replace_all_validate(replacements, reason)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 518, in replace_all_validate
    fgraph.replace(r, new_r, reason=reason, verbose=False)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/fg.py", line 486, in replace
    ". The type of the replacement must be the same.", old, new)
BadOptimization: BadOptimization Error 
  Variable: id 140341682289616 GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}.0
  Op GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}(GpuElemwise{sub,no_inplace}.0)
  Value Type: <type 'NoneType'>
  Old Value:  None
  New Value:  None
  Reason:  local_gpu_elemwise_careduce. The type of the replacement must be the same.
  Old Graph:
  GpuCAReduceCuda{add}{0, 2, 3} [id A] <GpuArrayType<None>(float32, vector)> ''   
   |GpuElemwise{sqr,no_inplace} [id B] <GpuArrayType<None>(float16, 4D)> ''   
     |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   
       |GpuDnnConv{algo='small', inplace=False} [id D] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuContiguous [id E] <GpuArrayType<None>(float16, 4D)> ''   
       | | |GpuElemwise{add,no_inplace} [id F] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuContiguous [id G] <GpuArrayType<None>(float16, 4D)> ''   
       | | |W [id H] <GpuArrayType<None>(float16, 4D)>
       | |GpuAllocEmpty{dtype='float16', context_name=None} [id I] <GpuArrayType<None>(float16, 4D)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[0] < 0).'} [id J] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[1] < 0).'} [id K] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[2] <= 0).'} [id L] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[3] <= 0).'} [id M] <TensorType(int64, scalar)> ''   
       | |GpuDnnConvDesc{border_mode='half', subsample=(1, 1), dilation=(1, 1), conv_mode='conv', precision='float32'} [id N] <CDataType{cudnnConvolutionDescriptor_t}> ''   
       | | |Shape [id O] <TensorType(int64, vector)> ''   
       | |Cast{float32} [id P] <float32> ''   
       | | |Constant{1.0} [id Q] <float64>
       | |Cast{float32} [id R] <float32> ''   
       |   |Constant{0.0} [id S] <float64>
       |GpuElemwise{Cast{float16}}[]<gpuarray> [id T] <GpuArrayType<None>(float16, (True, False, True, True))> ''   
         |GpuElemwise{true_div,no_inplace} [id U] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuElemwise{true_div,no_inplace} [id V] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuFromHost<None> [id W] <GpuArrayType<None>(float32, (True, True, True, True))> ''   

  New Graph:
  GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3} [id X] <GpuArrayType<None>(float16, vector)> ''   
   |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   


Hint: relax the tolerance by setting tensor.cmp_sloppy=1
  or even tensor.cmp_sloppy=2 for less-strict comparison


ERROR (theano.gof.opt): Optimization failure due to: local_gpu_elemwise_careduce
ERROR (theano.gof.opt): node: GpuCAReduceCuda{add}{0, 2, 3}(GpuElemwise{sqr,no_inplace}.0)
ERROR (theano.gof.opt): TRACEBACK:
ERROR (theano.gof.opt): Traceback (most recent call last):
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/opt.py", line 2036, in process_node
    remove=remove)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 569, in replace_all_validate_remove
    chk = fgraph.replace_all_validate(replacements, reason)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 518, in replace_all_validate
    fgraph.replace(r, new_r, reason=reason, verbose=False)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/fg.py", line 486, in replace
    ". The type of the replacement must be the same.", old, new)
BadOptimization: BadOptimization Error 
  Variable: id 140341682323216 GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}.0
  Op GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}(GpuElemwise{sub,no_inplace}.0)
  Value Type: <type 'NoneType'>
  Old Value:  None
  New Value:  None
  Reason:  local_gpu_elemwise_careduce. The type of the replacement must be the same.
  Old Graph:
  GpuCAReduceCuda{add}{0, 2, 3} [id A] <GpuArrayType<None>(float32, vector)> ''   
   |GpuElemwise{sqr,no_inplace} [id B] <GpuArrayType<None>(float16, 4D)> ''   
     |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   
       |GpuDnnConv{algo='small', inplace=False} [id D] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuContiguous [id E] <GpuArrayType<None>(float16, 4D)> ''   
       | | |GpuElemwise{add,no_inplace} [id F] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuContiguous [id G] <GpuArrayType<None>(float16, 4D)> ''   
       | | |W [id H] <GpuArrayType<None>(float16, 4D)>
       | |GpuAllocEmpty{dtype='float16', context_name=None} [id I] <GpuArrayType<None>(float16, 4D)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[0] < 0).'} [id J] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[1] < 0).'} [id K] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[2] <= 0).'} [id L] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[3] <= 0).'} [id M] <TensorType(int64, scalar)> ''   
       | |GpuDnnConvDesc{border_mode='half', subsample=(2, 2), dilation=(1, 1), conv_mode='conv', precision='float32'} [id N] <CDataType{cudnnConvolutionDescriptor_t}> ''   
       | | |Shape [id O] <TensorType(int64, vector)> ''   
       | |Cast{float32} [id P] <float32> ''   
       | | |Constant{1.0} [id Q] <float64>
       | |Cast{float32} [id R] <float32> ''   
       |   |Constant{0.0} [id S] <float64>
       |GpuElemwise{Cast{float16}}[]<gpuarray> [id T] <GpuArrayType<None>(float16, (True, False, True, True))> ''   
         |GpuElemwise{true_div,no_inplace} [id U] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuElemwise{true_div,no_inplace} [id V] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuFromHost<None> [id W] <GpuArrayType<None>(float32, (True, True, True, True))> ''   

  New Graph:
  GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3} [id X] <GpuArrayType<None>(float16, vector)> ''   
   |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   


Hint: relax the tolerance by setting tensor.cmp_sloppy=1
  or even tensor.cmp_sloppy=2 for less-strict comparison


ERROR (theano.gof.opt): Optimization failure due to: local_gpu_elemwise_careduce
ERROR (theano.gof.opt): node: GpuCAReduceCuda{add}{0, 2, 3}(GpuElemwise{sqr,no_inplace}.0)
ERROR (theano.gof.opt): TRACEBACK:
ERROR (theano.gof.opt): Traceback (most recent call last):
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/opt.py", line 2036, in process_node
    remove=remove)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 569, in replace_all_validate_remove
    chk = fgraph.replace_all_validate(replacements, reason)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 518, in replace_all_validate
    fgraph.replace(r, new_r, reason=reason, verbose=False)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/fg.py", line 486, in replace
    ". The type of the replacement must be the same.", old, new)
BadOptimization: BadOptimization Error 
  Variable: id 140341629247824 GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}.0
  Op GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}(GpuElemwise{sub,no_inplace}.0)
  Value Type: <type 'NoneType'>
  Old Value:  None
  New Value:  None
  Reason:  local_gpu_elemwise_careduce. The type of the replacement must be the same.
  Old Graph:
  GpuCAReduceCuda{add}{0, 2, 3} [id A] <GpuArrayType<None>(float32, vector)> ''   
   |GpuElemwise{sqr,no_inplace} [id B] <GpuArrayType<None>(float16, 4D)> ''   
     |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   
       |GpuDnnConv{algo='small', inplace=False} [id D] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuContiguous [id E] <GpuArrayType<None>(float16, 4D)> ''   
       | | |GpuFromHost<None> [id F] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuContiguous [id G] <GpuArrayType<None>(float16, 4D)> ''   
       | | |W [id H] <GpuArrayType<None>(float16, 4D)>
       | |GpuAllocEmpty{dtype='float16', context_name=None} [id I] <GpuArrayType<None>(float16, 4D)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[0] < 0).'} [id J] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[1] < 0).'} [id K] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[2] <= 0).'} [id L] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[3] <= 0).'} [id M] <TensorType(int64, scalar)> ''   
       | |GpuDnnConvDesc{border_mode='half', subsample=(1, 1), dilation=(1, 1), conv_mode='conv', precision='float32'} [id N] <CDataType{cudnnConvolutionDescriptor_t}> ''   
       | | |Shape [id O] <TensorType(int64, vector)> ''   
       | |Cast{float32} [id P] <float32> ''   
       | | |Constant{1.0} [id Q] <float64>
       | |Cast{float32} [id R] <float32> ''   
       |   |Constant{0.0} [id S] <float64>
       |GpuElemwise{Cast{float16}}[]<gpuarray> [id T] <GpuArrayType<None>(float16, (True, False, True, True))> ''   
         |GpuElemwise{true_div,no_inplace} [id U] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuElemwise{true_div,no_inplace} [id V] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuFromHost<None> [id W] <GpuArrayType<None>(float32, (True, True, True, True))> ''   

  New Graph:
  GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3} [id X] <GpuArrayType<None>(float16, vector)> ''   
   |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   


Hint: relax the tolerance by setting tensor.cmp_sloppy=1
  or even tensor.cmp_sloppy=2 for less-strict comparison


ERROR (theano.gof.opt): Optimization failure due to: local_gpu_elemwise_careduce
ERROR (theano.gof.opt): node: GpuCAReduceCuda{add}{0, 2, 3}(GpuElemwise{sqr,no_inplace}.0)
ERROR (theano.gof.opt): TRACEBACK:
ERROR (theano.gof.opt): Traceback (most recent call last):
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/opt.py", line 2036, in process_node
    remove=remove)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 569, in replace_all_validate_remove
    chk = fgraph.replace_all_validate(replacements, reason)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 518, in replace_all_validate
    fgraph.replace(r, new_r, reason=reason, verbose=False)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/fg.py", line 486, in replace
    ". The type of the replacement must be the same.", old, new)
BadOptimization: BadOptimization Error 
  Variable: id 140341682289616 GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}.0
  Op GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}(GpuElemwise{sub,no_inplace}.0)
  Value Type: <type 'NoneType'>
  Old Value:  None
  New Value:  None
  Reason:  local_gpu_elemwise_careduce. The type of the replacement must be the same.
  Old Graph:
  GpuCAReduceCuda{add}{0, 2, 3} [id A] <GpuArrayType<None>(float32, vector)> ''   
   |GpuElemwise{sqr,no_inplace} [id B] <GpuArrayType<None>(float16, 4D)> ''   
     |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   
       |GpuDnnConv{algo='small', inplace=False} [id D] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuContiguous [id E] <GpuArrayType<None>(float16, 4D)> ''   
       | | |GpuElemwise{add,no_inplace} [id F] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuContiguous [id G] <GpuArrayType<None>(float16, 4D)> ''   
       | | |W [id H] <GpuArrayType<None>(float16, 4D)>
       | |GpuAllocEmpty{dtype='float16', context_name=None} [id I] <GpuArrayType<None>(float16, 4D)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[0] < 0).'} [id J] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[1] < 0).'} [id K] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[2] <= 0).'} [id L] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[3] <= 0).'} [id M] <TensorType(int64, scalar)> ''   
       | |GpuDnnConvDesc{border_mode='half', subsample=(2, 2), dilation=(1, 1), conv_mode='conv', precision='float32'} [id N] <CDataType{cudnnConvolutionDescriptor_t}> ''   
       | | |Shape [id O] <TensorType(int64, vector)> ''   
       | |Constant{1.0} [id P] <float32>
       | |Constant{0.0} [id Q] <float32>
       |GpuElemwise{Cast{float16}}[]<gpuarray> [id R] <GpuArrayType<None>(float16, (True, False, True, True))> ''   
         |GpuElemwise{true_div,no_inplace} [id S] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuElemwise{true_div,no_inplace} [id T] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuFromHost<None> [id U] <GpuArrayType<None>(float32, (True, True, True, True))> ''   

  New Graph:
  GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3} [id V] <GpuArrayType<None>(float16, vector)> ''   
   |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   


Hint: relax the tolerance by setting tensor.cmp_sloppy=1
  or even tensor.cmp_sloppy=2 for less-strict comparison


ERROR (theano.gof.opt): Optimization failure due to: local_gpu_elemwise_careduce
ERROR (theano.gof.opt): node: GpuCAReduceCuda{add}{0, 2, 3}(GpuElemwise{sqr,no_inplace}.0)
ERROR (theano.gof.opt): TRACEBACK:
ERROR (theano.gof.opt): Traceback (most recent call last):
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/opt.py", line 2036, in process_node
    remove=remove)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 569, in replace_all_validate_remove
    chk = fgraph.replace_all_validate(replacements, reason)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 518, in replace_all_validate
    fgraph.replace(r, new_r, reason=reason, verbose=False)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/fg.py", line 486, in replace
    ". The type of the replacement must be the same.", old, new)
BadOptimization: BadOptimization Error 
  Variable: id 140341631692304 GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}.0
  Op GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}(GpuElemwise{sub,no_inplace}.0)
  Value Type: <type 'NoneType'>
  Old Value:  None
  New Value:  None
  Reason:  local_gpu_elemwise_careduce. The type of the replacement must be the same.
  Old Graph:
  GpuCAReduceCuda{add}{0, 2, 3} [id A] <GpuArrayType<None>(float32, vector)> ''   
   |GpuElemwise{sqr,no_inplace} [id B] <GpuArrayType<None>(float16, 4D)> ''   
     |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   
       |GpuDnnConv{algo='small', inplace=False} [id D] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuContiguous [id E] <GpuArrayType<None>(float16, 4D)> ''   
       | | |GpuElemwise{add,no_inplace} [id F] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuContiguous [id G] <GpuArrayType<None>(float16, 4D)> ''   
       | | |W [id H] <GpuArrayType<None>(float16, 4D)>
       | |GpuAllocEmpty{dtype='float16', context_name=None} [id I] <GpuArrayType<None>(float16, 4D)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[0] < 0).'} [id J] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[1] < 0).'} [id K] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[2] <= 0).'} [id L] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[3] <= 0).'} [id M] <TensorType(int64, scalar)> ''   
       | |GpuDnnConvDesc{border_mode='half', subsample=(1, 1), dilation=(1, 1), conv_mode='conv', precision='float32'} [id N] <CDataType{cudnnConvolutionDescriptor_t}> ''   
       | | |Shape [id O] <TensorType(int64, vector)> ''   
       | |Constant{1.0} [id P] <float32>
       | |Constant{0.0} [id Q] <float32>
       |GpuElemwise{Cast{float16}}[]<gpuarray> [id R] <GpuArrayType<None>(float16, (True, False, True, True))> ''   
         |GpuElemwise{true_div,no_inplace} [id S] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuElemwise{true_div,no_inplace} [id T] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuFromHost<None> [id U] <GpuArrayType<None>(float32, (True, True, True, True))> ''   

  New Graph:
  GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3} [id V] <GpuArrayType<None>(float16, vector)> ''   
   |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   


Hint: relax the tolerance by setting tensor.cmp_sloppy=1
  or even tensor.cmp_sloppy=2 for less-strict comparison


ERROR (theano.gof.opt): Optimization failure due to: local_gpu_elemwise_careduce
ERROR (theano.gof.opt): node: GpuCAReduceCuda{add}{0, 2, 3}(GpuElemwise{sqr,no_inplace}.0)
ERROR (theano.gof.opt): TRACEBACK:
ERROR (theano.gof.opt): Traceback (most recent call last):
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/opt.py", line 2036, in process_node
    remove=remove)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 569, in replace_all_validate_remove
    chk = fgraph.replace_all_validate(replacements, reason)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 518, in replace_all_validate
    fgraph.replace(r, new_r, reason=reason, verbose=False)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/fg.py", line 486, in replace
    ". The type of the replacement must be the same.", old, new)
BadOptimization: BadOptimization Error 
  Variable: id 140341664723280 GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}.0
  Op GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}(GpuElemwise{sub,no_inplace}.0)
  Value Type: <type 'NoneType'>
  Old Value:  None
  New Value:  None
  Reason:  local_gpu_elemwise_careduce. The type of the replacement must be the same.
  Old Graph:
  GpuCAReduceCuda{add}{0, 2, 3} [id A] <GpuArrayType<None>(float32, vector)> ''   
   |GpuElemwise{sqr,no_inplace} [id B] <GpuArrayType<None>(float16, 4D)> ''   
     |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   
       |GpuDnnConv{algo='small', inplace=False} [id D] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuContiguous [id E] <GpuArrayType<None>(float16, 4D)> ''   
       | | |GpuElemwise{add,no_inplace} [id F] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuContiguous [id G] <GpuArrayType<None>(float16, 4D)> ''   
       | | |W [id H] <GpuArrayType<None>(float16, 4D)>
       | |GpuAllocEmpty{dtype='float16', context_name=None} [id I] <GpuArrayType<None>(float16, 4D)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[0] < 0).'} [id J] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[1] < 0).'} [id K] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[2] <= 0).'} [id L] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[3] <= 0).'} [id M] <TensorType(int64, scalar)> ''   
       | |GpuDnnConvDesc{border_mode='half', subsample=(2, 2), dilation=(1, 1), conv_mode='conv', precision='float32'} [id N] <CDataType{cudnnConvolutionDescriptor_t}> ''   
       | | |Shape [id O] <TensorType(int64, vector)> ''   
       | |Constant{1.0} [id P] <float32>
       | |Constant{0.0} [id Q] <float32>
       |GpuElemwise{Cast{float16}}[]<gpuarray> [id R] <GpuArrayType<None>(float16, (True, False, True, True))> ''   
         |GpuElemwise{true_div,no_inplace} [id S] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuElemwise{true_div,no_inplace} [id T] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuFromHost<None> [id U] <GpuArrayType<None>(float32, (True, True, True, True))> ''   

  New Graph:
  GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3} [id V] <GpuArrayType<None>(float16, vector)> ''   
   |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   


Hint: relax the tolerance by setting tensor.cmp_sloppy=1
  or even tensor.cmp_sloppy=2 for less-strict comparison


ERROR (theano.gof.opt): Optimization failure due to: local_gpu_elemwise_careduce
ERROR (theano.gof.opt): node: GpuCAReduceCuda{add}{0, 2, 3}(GpuElemwise{sqr,no_inplace}.0)
ERROR (theano.gof.opt): TRACEBACK:
ERROR (theano.gof.opt): Traceback (most recent call last):
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/opt.py", line 2036, in process_node
    remove=remove)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 569, in replace_all_validate_remove
    chk = fgraph.replace_all_validate(replacements, reason)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 518, in replace_all_validate
    fgraph.replace(r, new_r, reason=reason, verbose=False)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/fg.py", line 486, in replace
    ". The type of the replacement must be the same.", old, new)
BadOptimization: BadOptimization Error 
  Variable: id 140341665096400 GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}.0
  Op GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}(GpuElemwise{sub,no_inplace}.0)
  Value Type: <type 'NoneType'>
  Old Value:  None
  New Value:  None
  Reason:  local_gpu_elemwise_careduce. The type of the replacement must be the same.
  Old Graph:
  GpuCAReduceCuda{add}{0, 2, 3} [id A] <GpuArrayType<None>(float32, vector)> ''   
   |GpuElemwise{sqr,no_inplace} [id B] <GpuArrayType<None>(float16, 4D)> ''   
     |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   
       |GpuDnnConv{algo='small', inplace=False} [id D] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuContiguous [id E] <GpuArrayType<None>(float16, 4D)> ''   
       | | |GpuElemwise{add,no_inplace} [id F] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuContiguous [id G] <GpuArrayType<None>(float16, 4D)> ''   
       | | |W [id H] <GpuArrayType<None>(float16, 4D)>
       | |GpuAllocEmpty{dtype='float16', context_name=None} [id I] <GpuArrayType<None>(float16, 4D)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[0] < 0).'} [id J] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[1] < 0).'} [id K] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[2] <= 0).'} [id L] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[3] <= 0).'} [id M] <TensorType(int64, scalar)> ''   
       | |GpuDnnConvDesc{border_mode='half', subsample=(1, 1), dilation=(1, 1), conv_mode='conv', precision='float32'} [id N] <CDataType{cudnnConvolutionDescriptor_t}> ''   
       | | |Shape [id O] <TensorType(int64, vector)> ''   
       | |Constant{1.0} [id P] <float32>
       | |Constant{0.0} [id Q] <float32>
       |GpuElemwise{Cast{float16}}[]<gpuarray> [id R] <GpuArrayType<None>(float16, (True, False, True, True))> ''   
         |GpuElemwise{true_div,no_inplace} [id S] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuElemwise{true_div,no_inplace} [id T] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuFromHost<None> [id U] <GpuArrayType<None>(float32, (True, True, True, True))> ''   

  New Graph:
  GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3} [id V] <GpuArrayType<None>(float16, vector)> ''   
   |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   


Hint: relax the tolerance by setting tensor.cmp_sloppy=1
  or even tensor.cmp_sloppy=2 for less-strict comparison


ERROR (theano.gof.opt): Optimization failure due to: local_gpu_elemwise_careduce
ERROR (theano.gof.opt): node: GpuCAReduceCuda{add}{0, 2, 3}(GpuElemwise{sqr,no_inplace}.0)
ERROR (theano.gof.opt): TRACEBACK:
ERROR (theano.gof.opt): Traceback (most recent call last):
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/opt.py", line 2036, in process_node
    remove=remove)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 569, in replace_all_validate_remove
    chk = fgraph.replace_all_validate(replacements, reason)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 518, in replace_all_validate
    fgraph.replace(r, new_r, reason=reason, verbose=False)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/fg.py", line 486, in replace
    ". The type of the replacement must be the same.", old, new)
BadOptimization: BadOptimization Error 
  Variable: id 140341682752656 GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}.0
  Op GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}(GpuElemwise{sub,no_inplace}.0)
  Value Type: <type 'NoneType'>
  Old Value:  None
  New Value:  None
  Reason:  local_gpu_elemwise_careduce. The type of the replacement must be the same.
  Old Graph:
  GpuCAReduceCuda{add}{0, 2, 3} [id A] <GpuArrayType<None>(float32, vector)> ''   
   |GpuElemwise{sqr,no_inplace} [id B] <GpuArrayType<None>(float16, 4D)> ''   
     |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   
       |GpuDnnConv{algo='small', inplace=False} [id D] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuContiguous [id E] <GpuArrayType<None>(float16, 4D)> ''   
       | | |GpuElemwise{add,no_inplace} [id F] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuContiguous [id G] <GpuArrayType<None>(float16, 4D)> ''   
       | | |W [id H] <GpuArrayType<None>(float16, 4D)>
       | |GpuAllocEmpty{dtype='float16', context_name=None} [id I] <GpuArrayType<None>(float16, 4D)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[0] < 0).'} [id J] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[1] < 0).'} [id K] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[2] <= 0).'} [id L] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[3] <= 0).'} [id M] <TensorType(int64, scalar)> ''   
       | |GpuDnnConvDesc{border_mode='half', subsample=(2, 2), dilation=(1, 1), conv_mode='conv', precision='float32'} [id N] <CDataType{cudnnConvolutionDescriptor_t}> ''   
       | | |Shape [id O] <TensorType(int64, vector)> ''   
       | |Constant{1.0} [id P] <float32>
       | |Constant{0.0} [id Q] <float32>
       |GpuElemwise{Cast{float16}}[]<gpuarray> [id R] <GpuArrayType<None>(float16, (True, False, True, True))> ''   
         |GpuElemwise{true_div,no_inplace} [id S] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuElemwise{true_div,no_inplace} [id T] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuFromHost<None> [id U] <GpuArrayType<None>(float32, (True, True, True, True))> ''   

  New Graph:
  GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3} [id V] <GpuArrayType<None>(float16, vector)> ''   
   |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   


Hint: relax the tolerance by setting tensor.cmp_sloppy=1
  or even tensor.cmp_sloppy=2 for less-strict comparison


ERROR (theano.gof.opt): Optimization failure due to: local_gpu_elemwise_careduce
ERROR (theano.gof.opt): node: GpuCAReduceCuda{add}{0, 2, 3}(GpuElemwise{sqr,no_inplace}.0)
ERROR (theano.gof.opt): TRACEBACK:
ERROR (theano.gof.opt): Traceback (most recent call last):
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/opt.py", line 2036, in process_node
    remove=remove)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 569, in replace_all_validate_remove
    chk = fgraph.replace_all_validate(replacements, reason)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 518, in replace_all_validate
    fgraph.replace(r, new_r, reason=reason, verbose=False)
  File "/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/fg.py", line 486, in replace
    ". The type of the replacement must be the same.", old, new)
BadOptimization: BadOptimization Error 
  Variable: id 140341682332624 GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}.0
  Op GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3}(GpuElemwise{sub,no_inplace}.0)
  Value Type: <type 'NoneType'>
  Old Value:  None
  New Value:  None
  Reason:  local_gpu_elemwise_careduce. The type of the replacement must be the same.
  Old Graph:
  GpuCAReduceCuda{add}{0, 2, 3} [id A] <GpuArrayType<None>(float32, vector)> ''   
   |GpuElemwise{sqr,no_inplace} [id B] <GpuArrayType<None>(float16, 4D)> ''   
     |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   
       |GpuDnnConv{algo='small', inplace=False} [id D] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuContiguous [id E] <GpuArrayType<None>(float16, 4D)> ''   
       | | |GpuFromHost<None> [id F] <GpuArrayType<None>(float16, 4D)> ''   
       | |GpuContiguous [id G] <GpuArrayType<None>(float16, 4D)> ''   
       | | |W [id H] <GpuArrayType<None>(float16, 4D)>
       | |GpuAllocEmpty{dtype='float16', context_name=None} [id I] <GpuArrayType<None>(float16, 4D)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[0] < 0).'} [id J] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[1] < 0).'} [id K] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[2] <= 0).'} [id L] <TensorType(int64, scalar)> ''   
       | | |Assert{msg='The convolution would produce an invalid shape (dim[3] <= 0).'} [id M] <TensorType(int64, scalar)> ''   
       | |GpuDnnConvDesc{border_mode='half', subsample=(1, 1), dilation=(1, 1), conv_mode='conv', precision='float32'} [id N] <CDataType{cudnnConvolutionDescriptor_t}> ''   
       | | |Shape [id O] <TensorType(int64, vector)> ''   
       | |Constant{1.0} [id P] <float32>
       | |Constant{0.0} [id Q] <float32>
       |GpuElemwise{Cast{float16}}[]<gpuarray> [id R] <GpuArrayType<None>(float16, (True, False, True, True))> ''   
         |GpuElemwise{true_div,no_inplace} [id S] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuElemwise{true_div,no_inplace} [id T] <GpuArrayType<None>(float32, (True, False, True, True))> ''   
           |GpuFromHost<None> [id U] <GpuArrayType<None>(float32, (True, True, True, True))> ''   

  New Graph:
  GpuCAReduceCuda{pre=sqr,red=add}{0, 2, 3} [id V] <GpuArrayType<None>(float16, vector)> ''   
   |GpuElemwise{sub,no_inplace} [id C] <GpuArrayType<None>(float16, 4D)> ''   


Hint: relax the tolerance by setting tensor.cmp_sloppy=1
  or even tensor.cmp_sloppy=2 for less-strict comparison


Disabling C code for Argmax due to unsupported float16
Disabling C code for Argmax due to unsupported float16
---------------------------------------------------------------------------
AssertionError                            Traceback (most recent call last)
<ipython-input-1-d5bab382d762> in <module>()
     86     for _ in range(10):
     87         d, t = gen_tr.next()
---> 88         l, a = train_fn(d, t)
     89         acc_tr.append(a)
     90         loss_tr.append(l)

/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/compile/function_module.pyc in __call__(self, *args, **kwargs)
    882         try:
    883             outputs =\
--> 884                 self.fn() if output_subset is None else\
    885                 self.fn(output_subset=output_subset)
    886         except Exception:

/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/vm.pyc in __call__(self, output_subset)
    517                             current_apply,
    518                             self.thunks[self.node_idx[current_apply]],
--> 519                             storage_map=storage_map)
    520                     for o in current_apply.outputs:
    521                         compute_map[o][0] = 1

/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/link.pyc in raise_with_op(node, thunk, exc_info, storage_map)
    323         # extra long error message in that case.
    324         pass
--> 325     reraise(exc_type, exc_value, exc_trace)
    326 
    327 

/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/vm.pyc in __call__(self, output_subset)
    484                     # -- Non-lazy case: have inputs, time to compute outputs
    485                     try:
--> 486                         _, dt = self.run_thunk_of_node(current_apply)
    487                         del _
    488                         if config.profile or config.print_global_stats:

/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/gof/vm.pyc in run_thunk_of_node(self, node)
    402                 thunk=self.thunks[idx],
    403                 storage_map=self.storage_map,
--> 404                 compute_map=self.compute_map,
    405             )
    406         return rval, dt

/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/compile/nanguardmode.pyc in nan_check(node, thunk, storage_map, compute_map)
    271                 if (compute_map[var][0] and
    272                         getattr(var.tag, 'nan_guard_mode_check', True)):
--> 273                     do_check_on(storage_map[var][0], node)
    274 
    275         def nan_check_input(var, value):

/home/fabian/deeplearning_venv/local/lib/python2.7/site-packages/theano/compile/nanguardmode.pyc in do_check_on(value, nd, var)
    259                 msg = sio.getvalue()
    260                 if config.NanGuardMode.action == 'raise':
--> 261                     raise AssertionError(msg)
    262                 elif config.NanGuardMode.action == 'pdb':
    263                     print(msg)

AssertionError: Inf detected
Big value detected
NanGuardMode found an error in the output of a node in this variable:
GpuElemwise{Composite{(i0 - ((i1 * i2) / sqrt(i3)))}}[(0, 0)]<gpuarray> [id A] ''   
 |W [id B]
 |InplaceGpuDimShuffle{x,x} [id C] ''   
 | |GpuElemwise{Composite{((i0 * sqrt((i1 - (i2 ** i3)))) / (i4 - (i5 ** i6)))}}[]<gpuarray> [id D] ''   
 |   |GpuArrayConstant{0.00500106811523} [id E]
 |   |GpuArrayConstant{1.0} [id F]
 |   |GpuArrayConstant{0.9990234375} [id G]
 |   |GpuFromHost<None> [id H] ''   
 |   | |Elemwise{add,no_inplace} [id I] ''   
 |   |   |TensorConstant{1.0} [id J]
 |   |   |<TensorType(float16, scalar)> [id K]
 |   |GpuArrayConstant{1.0} [id F]
 |   |GpuArrayConstant{0.89990234375} [id L]
 |   |GpuFromHost<None> [id H] ''   
 |GpuGemm{inplace=True} [id M] ''   
 | |GpuElemwise{Mul}[(0, 1)]<gpuarray> [id N] ''   
 | | |GpuArrayConstant{[[ 0.89990234]]} [id O]
 | | |<GpuArrayType<None>(float16, matrix)> [id P]
 | |TensorConstant{0.10009765625} [id Q]
 | |InplaceGpuDimShuffle{1,0} [id R] ''   
 | | |GpuElemwise{Composite{Cast{float16}((i0 / i1))}}[]<gpuarray> [id S] ''   
 | |   |GpuCAReduceCuda{add}{2} [id T] ''   
 | |   | |GpuReshape{3} [id U] ''   
 | |   |   |GpuElemwise{Composite{((i0 * i1) + (i2 * Abs(i1)))}}[]<gpuarray> [id V] ''   
 | |   |   | |GpuArrayConstant{[[[[ 0.50488281]]]]} [id W]
 | |   |   | |GpuElemwise{Composite{((i0 * i1 * i2) + i3)}}[]<gpuarray> [id X] ''   
 | |   |   | | |GpuElemwise{sub,no_inplace} [id Y] ''   
 | |   |   | | | |GpuDnnConv{algo='small', inplace=True} [id Z] ''   
 | |   |   | | | | |GpuContiguous [id BA] ''   
 | |   |   | | | | | |GpuElemwise{Composite{((i0 * i1) + (i2 * Abs(i1)))}}[]<gpuarray> [id BB] ''   
 | |   |   | | | | |   |GpuArrayConstant{[[[[ 0.50488281]]]]} [id W]
 | |   |   | | | | |   |GpuElemwise{Composite{((i0 * i1 * i2) + i3)}}[]<gpuarray> [id BC] ''   
 | |   |   | | | | |   | |GpuElemwise{Sub}[(0, 0)]<gpuarray> [id BD] ''   
 | |   |   | | | | |   | | |GpuDnnConv{algo='small', inplace=True} [id BE] ''   
 | |   |   | | | | |   | | | |GpuContiguous [id BF] ''   
 | |   |   | | | | |   | | | | |GpuElemwise{Composite{((i0 * i1) + (i2 * Abs(i1)))}}[]<gpuarray> [id BG] ''   
 | |   |   | | | | |   | | | |   |GpuArrayConstant{[[[[ 0.50488281]]]]} [id W]
 | |   |   | | | | |   | | | |   |GpuElemwise{Composite{((i0 * i1 * i2) + i3)}}[]<gpuarray> [id BH] ''   
 | |   |   | | | | |   | | | |   | |GpuElemwise{sub,no_inplace} [id BI] ''   
 | |   |   | | | | |   | | | |   | | |GpuDnnConv{algo='small', inplace=True} [id BJ] ''   
 | |   |   | | | | |   | | | |   | | | |GpuContiguous [id BK] ''   
 | |   |   | | | | |   | | | |   | | | | |GpuElemwise{Composite{((i0 * i1) + (i2 * Abs(i1)))}}[]<gpuarray> [id BL] ''   
 | |   |   | | | | |   | | | |   | | | |   |GpuArrayConstant{[[[[ 0.50488281]]]]} [id W]
 | |   |   | | | | |   | | | |   | | | |   |GpuElemwise{Composite{((i0 * i1 * i2) + i3)}}[]<gpuarray> [id BM] ''   
 | |   |   | | | | |   | | | |   | | | |   | |GpuElemwise{sub,no_inplace} [id BN] ''   
 | |   |   | | | | |   | | | |   | | | |   | | |GpuDnnConv{algo='small', inplace=True} [id BO] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |GpuContiguous [id BP] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | | |GpuElemwise{Composite{((i0 * i1) + (i2 * Abs(i1)))}}[]<gpuarray> [id BQ] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   |GpuArrayConstant{[[[[ 0.50488281]]]]} [id W]
 | |   |   | | | | |   | | | |   | | | |   | | | |   |GpuElemwise{Composite{((i0 * i1 * i2) + i3)}}[]<gpuarray> [id BR] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |GpuElemwise{sub,no_inplace} [id BS] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | |GpuDnnConv{algo='small', inplace=True} [id BT] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |GpuContiguous [id BU] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | |GpuElemwise{Composite{((i0 * i1) + (i2 * Abs(i1)))}}[]<gpuarray> [id BV] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   |GpuArrayConstant{[[[[ 0.50488281]]]]} [id W]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   |GpuElemwise{Composite{((i0 * i1 * i2) + i3)}}[]<gpuarray> [id BW] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |GpuElemwise{sub,no_inplace} [id BX] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | |GpuDnnConv{algo='small', inplace=True} [id BY] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | |GpuContiguous [id BZ] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | |GpuFromHost<None> [id CA] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | |   |<TensorType(float16, 4D)> [id CB]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | |GpuContiguous [id CC] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | |W [id CD]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | |GpuAllocEmpty{dtype='float16', context_name=None} [id CE] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | |Assert{msg='The convolution would produce an invalid shape (dim[0] < 0).'} [id CF] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | | |Shape_i{0} [id CG] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | | | |GpuContiguous [id BZ] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | | |Elemwise{ge,no_inplace} [id CH] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | |   |Shape_i{0} [id CG] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | |   |TensorConstant{0} [id CI]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | |Assert{msg='The convolution would produce an invalid shape (dim[1] < 0).'} [id CJ] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | | |Shape_i{0} [id CK] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | | | |GpuContiguous [id CC] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | | |Elemwise{ge,no_inplace} [id CL] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | |   |Shape_i{0} [id CK] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | |   |TensorConstant{0} [id CI]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | |Assert{msg='The convolution would produce an invalid shape (dim[2] <= 0).'} [id CM] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | | |Elemwise{Composite{(((i0 + (i1 * (Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5) // i6))) - Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5)) + i7)}} [id CN] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | | | |Shape_i{2} [id CO] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | | | | |GpuContiguous [id BZ] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | | | |TensorConstant{2} [id CP]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | | | |Shape_i{2} [id CQ] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | | | | |GpuContiguous [id CC] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | | | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | | | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | | | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | | | |TensorConstant{2} [id CP]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | | | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | | |Elemwise{gt,no_inplace} [id CS] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | |   |Elemwise{Composite{(((i0 + (i1 * (Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5) // i6))) - Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5)) + i7)}} [id CN] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | |   |TensorConstant{0} [id CI]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | |Assert{msg='The convolution would produce an invalid shape (dim[3] <= 0).'} [id CT] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | |   |Elemwise{Composite{(((i0 + (i1 * (Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5) // i6))) - Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5)) + i7)}} [id CU] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | |   | |Shape_i{3} [id CV] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | |   | | |GpuContiguous [id BZ] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | |   | |TensorConstant{2} [id CP]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | |   | |Shape_i{3} [id CW] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | |   | | |GpuContiguous [id CC] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | |   | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | |   | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | |   | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | |   | |TensorConstant{2} [id CP]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | |   | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | |   |Elemwise{gt,no_inplace} [id CX] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | |     |Elemwise{Composite{(((i0 + (i1 * (Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5) // i6))) - Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5)) + i7)}} [id CU] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | |     |TensorConstant{0} [id CI]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | |GpuDnnConvDesc{border_mode='half', subsample=(1, 1), dilation=(1, 1), conv_mode='conv', precision='float32'} [id CY] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | | |Shape [id CZ] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | |   |GpuContiguous [id CC] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | |Constant{1.0} [id DA]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | | |Constant{0.0} [id DB]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | |InplaceGpuDimShuffle{x,0,x,x} [id DC] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |   |GpuElemwise{Composite{Cast{float16}((((i0 / i1) / i2) / i3))}}[]<gpuarray> [id DD] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |GpuCAReduceCuda{add}{0, 2, 3} [id DE] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     | |GpuDnnConv{algo='small', inplace=True} [id BY] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |InplaceGpuDimShuffle{x} [id DF] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     | |GpuSubtensor{int64} [id DG] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |   |GpuElemwise{Cast{float32}}[]<gpuarray> [id DH] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |   | |GpuFromHost<None> [id DI] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |   |   |MakeVector{dtype='int64'} [id DJ] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |   |     |Shape_i{0} [id DK] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |   |     | |<TensorType(float16, 4D)> [id CB]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |   |     |TensorConstant{32} [id DL]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |   |     |TensorConstant{32} [id DL]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |   |     |TensorConstant{32} [id DL]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |   |Constant{0} [id DM]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |InplaceGpuDimShuffle{x} [id DN] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     | |GpuSubtensor{int64} [id DO] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |   |GpuElemwise{Cast{float32}}[]<gpuarray> [id DH] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |   |Constant{2} [id DP]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |InplaceGpuDimShuffle{x} [id DQ] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |       |GpuSubtensor{int64} [id DR] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |         |GpuElemwise{Cast{float32}}[]<gpuarray> [id DH] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |         |Constant{3} [id DS]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |InplaceGpuDimShuffle{x,0,x,x} [id DT] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | |gamma [id DU]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |InplaceGpuDimShuffle{x,0,x,x} [id DV] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | |GpuElemwise{inv,no_inplace} [id DW] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |   |GpuElemwise{Composite{sqrt((i0 + Cast{float16}((((i1 / i2) / i3) / i4))))}}[]<gpuarray> [id DX] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |GpuArrayConstant{[ 0.00010002]} [id DY]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |GpuCAReduceCuda{add}{0, 2, 3} [id DZ] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     | |GpuElemwise{sqr,no_inplace} [id EA] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |   |GpuElemwise{sub,no_inplace} [id EB] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |     |GpuDnnConv{algo='small', inplace=True} [id BY] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |     |GpuElemwise{Composite{Cast{float16}((((i0 / i1) / i2) / i3))}}[]<gpuarray> [id EC] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |       |InplaceGpuDimShuffle{x,0,x,x} [id ED] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |       | |GpuCAReduceCuda{add}{0, 2, 3} [id DE] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |       |InplaceGpuDimShuffle{x,x,x,x} [id EE] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |       | |GpuSubtensor{int64} [id DG] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |       |InplaceGpuDimShuffle{x,x,x,x} [id EF] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |       | |GpuSubtensor{int64} [id DO] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |       |InplaceGpuDimShuffle{x,x,x,x} [id EG] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |         |GpuSubtensor{int64} [id DR] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |InplaceGpuDimShuffle{x} [id DF] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |InplaceGpuDimShuffle{x} [id DN] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |     |InplaceGpuDimShuffle{x} [id DQ] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |InplaceGpuDimShuffle{x,0,x,x} [id EH] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   |   |beta [id EI]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   |GpuArrayConstant{[[[[ 0.49511719]]]]} [id EJ]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |GpuContiguous [id EK] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | |W [id EL]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |GpuAllocEmpty{dtype='float16', context_name=None} [id EM] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | |Assert{msg='The convolution would produce an invalid shape (dim[0] < 0).'} [id EN] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | | |Shape_i{0} [id EO] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | | | |GpuContiguous [id BU] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | | |Elemwise{ge,no_inplace} [id EP] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | |   |Shape_i{0} [id EO] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | |   |TensorConstant{0} [id CI]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | |Assert{msg='The convolution would produce an invalid shape (dim[1] < 0).'} [id EQ] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | | |Shape_i{0} [id ER] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | | | |GpuContiguous [id EK] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | | |Elemwise{ge,no_inplace} [id ES] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | |   |Shape_i{0} [id ER] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | |   |TensorConstant{0} [id CI]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | |Assert{msg='The convolution would produce an invalid shape (dim[2] <= 0).'} [id ET] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | | |Elemwise{Composite{((((i0 + (i1 * (Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5) // i6))) - Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5)) // i7) + i8)}} [id EU] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | | | |Shape_i{2} [id EV] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | | | | |GpuContiguous [id BU] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | | | |TensorConstant{2} [id CP]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | | | |Shape_i{2} [id EW] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | | | | |GpuContiguous [id EK] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | | | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | | | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | | | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | | | |TensorConstant{2} [id CP]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | | | |TensorConstant{2} [id CP]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | | | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | | |Elemwise{gt,no_inplace} [id EX] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | |   |Elemwise{Composite{((((i0 + (i1 * (Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5) // i6))) - Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5)) // i7) + i8)}} [id EU] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | |   |TensorConstant{0} [id CI]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | |Assert{msg='The convolution would produce an invalid shape (dim[3] <= 0).'} [id EY] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   |Elemwise{Composite{((((i0 + (i1 * (Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5) // i6))) - Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5)) // i7) + i8)}} [id EZ] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |Shape_i{3} [id FA] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | |GpuContiguous [id BU] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |TensorConstant{2} [id CP]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |Shape_i{3} [id FB] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | | |GpuContiguous [id EK] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |TensorConstant{2} [id CP]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |TensorConstant{2} [id CP]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   |Elemwise{gt,no_inplace} [id FC] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |     |Elemwise{Composite{((((i0 + (i1 * (Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5) // i6))) - Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5)) // i7) + i8)}} [id EZ] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |     |TensorConstant{0} [id CI]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |GpuDnnConvDesc{border_mode='half', subsample=(2, 2), dilation=(1, 1), conv_mode='conv', precision='float32'} [id FD] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | | |Shape [id FE] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |   |GpuContiguous [id EK] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |Constant{1.0} [id DA]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | | |Constant{0.0} [id DB]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | |InplaceGpuDimShuffle{x,0,x,x} [id FF] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |   |GpuElemwise{Composite{Cast{float16}((((i0 / i1) / i2) / i3))}}[]<gpuarray> [id FG] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |GpuCAReduceCuda{add}{0, 2, 3} [id FH] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     | |GpuDnnConv{algo='small', inplace=True} [id BT] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |InplaceGpuDimShuffle{x} [id FI] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     | |GpuSubtensor{int64} [id FJ] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |   |GpuElemwise{Cast{float32}}[]<gpuarray> [id FK] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |   | |GpuFromHost<None> [id FL] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |   |   |MakeVector{dtype='int64'} [id FM] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |   |     |Shape_i{0} [id DK] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |   |     |TensorConstant{32} [id DL]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |   |     |TensorConstant{16} [id FN]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |   |     |TensorConstant{16} [id FN]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |   |Constant{0} [id DM]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |InplaceGpuDimShuffle{x} [id FO] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     | |GpuSubtensor{int64} [id FP] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |   |GpuElemwise{Cast{float32}}[]<gpuarray> [id FK] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |   |Constant{2} [id DP]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |InplaceGpuDimShuffle{x} [id FQ] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |       |GpuSubtensor{int64} [id FR] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |         |GpuElemwise{Cast{float32}}[]<gpuarray> [id FK] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |         |Constant{3} [id DS]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |InplaceGpuDimShuffle{x,0,x,x} [id FS] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | |gamma [id FT]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |InplaceGpuDimShuffle{x,0,x,x} [id FU] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | |GpuElemwise{inv,no_inplace} [id FV] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |   |GpuElemwise{Composite{sqrt((i0 + Cast{float16}((((i1 / i2) / i3) / i4))))}}[]<gpuarray> [id FW] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |GpuArrayConstant{[ 0.00010002]} [id DY]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |GpuCAReduceCuda{add}{0, 2, 3} [id FX] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     | |GpuElemwise{sqr,no_inplace} [id FY] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |   |GpuElemwise{sub,no_inplace} [id FZ] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |     |GpuDnnConv{algo='small', inplace=True} [id BT] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |     |GpuElemwise{Composite{Cast{float16}((((i0 / i1) / i2) / i3))}}[]<gpuarray> [id GA] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |       |InplaceGpuDimShuffle{x,0,x,x} [id GB] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |       | |GpuCAReduceCuda{add}{0, 2, 3} [id FH] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |       |InplaceGpuDimShuffle{x,x,x,x} [id GC] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |       | |GpuSubtensor{int64} [id FJ] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |       |InplaceGpuDimShuffle{x,x,x,x} [id GD] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |       | |GpuSubtensor{int64} [id FP] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |       |InplaceGpuDimShuffle{x,x,x,x} [id GE] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |         |GpuSubtensor{int64} [id FR] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |InplaceGpuDimShuffle{x} [id FI] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |InplaceGpuDimShuffle{x} [id FO] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |     |InplaceGpuDimShuffle{x} [id FQ] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |InplaceGpuDimShuffle{x,0,x,x} [id GF] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   |   |beta [id GG]
 | |   |   | | | | |   | | | |   | | | |   | | | |   |GpuArrayConstant{[[[[ 0.49511719]]]]} [id EJ]
 | |   |   | | | | |   | | | |   | | | |   | | | |GpuContiguous [id GH] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | | |W [id GI]
 | |   |   | | | | |   | | | |   | | | |   | | | |GpuAllocEmpty{dtype='float16', context_name=None} [id GJ] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | | |Assert{msg='The convolution would produce an invalid shape (dim[0] < 0).'} [id GK] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | | | |Shape_i{0} [id GL] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | | | | |GpuContiguous [id BP] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | | | |Elemwise{ge,no_inplace} [id GM] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | | |   |Shape_i{0} [id GL] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | | |   |TensorConstant{0} [id CI]
 | |   |   | | | | |   | | | |   | | | |   | | | | |Assert{msg='The convolution would produce an invalid shape (dim[1] < 0).'} [id GN] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | | | |Shape_i{0} [id GO] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | | | | |GpuContiguous [id GH] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | | | |Elemwise{ge,no_inplace} [id GP] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | | |   |Shape_i{0} [id GO] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | | |   |TensorConstant{0} [id CI]
 | |   |   | | | | |   | | | |   | | | |   | | | | |Assert{msg='The convolution would produce an invalid shape (dim[2] <= 0).'} [id GQ] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | | | |Elemwise{Composite{(((i0 + (i1 * (Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5) // i6))) - Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5)) + i7)}} [id GR] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | | | | |Shape_i{2} [id GS] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | | | | | |GpuContiguous [id BP] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | | | | |TensorConstant{2} [id CP]
 | |   |   | | | | |   | | | |   | | | |   | | | | | | |Shape_i{2} [id GT] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | | | | | |GpuContiguous [id GH] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | | | | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | |   | | | | | | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | |   | | | | | | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | |   | | | | | | |TensorConstant{2} [id CP]
 | |   |   | | | | |   | | | |   | | | |   | | | | | | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | |   | | | | | |Elemwise{gt,no_inplace} [id GU] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | | |   |Elemwise{Composite{(((i0 + (i1 * (Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5) // i6))) - Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5)) + i7)}} [id GR] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | | |   |TensorConstant{0} [id CI]
 | |   |   | | | | |   | | | |   | | | |   | | | | |Assert{msg='The convolution would produce an invalid shape (dim[3] <= 0).'} [id GV] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   |Elemwise{Composite{(((i0 + (i1 * (Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5) // i6))) - Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5)) + i7)}} [id GW] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |Shape_i{3} [id GX] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | |GpuContiguous [id BP] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |TensorConstant{2} [id CP]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |Shape_i{3} [id GY] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | | |GpuContiguous [id GH] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |TensorConstant{2} [id CP]
 | |   |   | | | | |   | | | |   | | | |   | | | |   | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | |   | | | |   |Elemwise{gt,no_inplace} [id GZ] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |     |Elemwise{Composite{(((i0 + (i1 * (Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5) // i6))) - Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5)) + i7)}} [id GW] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |     |TensorConstant{0} [id CI]
 | |   |   | | | | |   | | | |   | | | |   | | | |GpuDnnConvDesc{border_mode='half', subsample=(1, 1), dilation=(1, 1), conv_mode='conv', precision='float32'} [id HA] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | | |Shape [id HB] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |   |GpuContiguous [id GH] ''   
 | |   |   | | | | |   | | | |   | | | |   | | | |Constant{1.0} [id DA]
 | |   |   | | | | |   | | | |   | | | |   | | | |Constant{0.0} [id DB]
 | |   |   | | | | |   | | | |   | | | |   | | |InplaceGpuDimShuffle{x,0,x,x} [id HC] ''   
 | |   |   | | | | |   | | | |   | | | |   | |   |GpuElemwise{Composite{Cast{float16}((((i0 / i1) / i2) / i3))}}[]<gpuarray> [id HD] ''   
 | |   |   | | | | |   | | | |   | | | |   | |     |GpuCAReduceCuda{add}{0, 2, 3} [id HE] ''   
 | |   |   | | | | |   | | | |   | | | |   | |     | |GpuDnnConv{algo='small', inplace=True} [id BO] ''   
 | |   |   | | | | |   | | | |   | | | |   | |     |InplaceGpuDimShuffle{x} [id HF] ''   
 | |   |   | | | | |   | | | |   | | | |   | |     | |GpuSubtensor{int64} [id HG] ''   
 | |   |   | | | | |   | | | |   | | | |   | |     |   |GpuElemwise{Cast{float32}}[]<gpuarray> [id HH] ''   
 | |   |   | | | | |   | | | |   | | | |   | |     |   | |GpuFromHost<None> [id HI] ''   
 | |   |   | | | | |   | | | |   | | | |   | |     |   |   |MakeVector{dtype='int64'} [id HJ] ''   
 | |   |   | | | | |   | | | |   | | | |   | |     |   |     |Shape_i{0} [id DK] ''   
 | |   |   | | | | |   | | | |   | | | |   | |     |   |     |TensorConstant{64} [id HK]
 | |   |   | | | | |   | | | |   | | | |   | |     |   |     |TensorConstant{16} [id FN]
 | |   |   | | | | |   | | | |   | | | |   | |     |   |     |TensorConstant{16} [id FN]
 | |   |   | | | | |   | | | |   | | | |   | |     |   |Constant{0} [id DM]
 | |   |   | | | | |   | | | |   | | | |   | |     |InplaceGpuDimShuffle{x} [id HL] ''   
 | |   |   | | | | |   | | | |   | | | |   | |     | |GpuSubtensor{int64} [id HM] ''   
 | |   |   | | | | |   | | | |   | | | |   | |     |   |GpuElemwise{Cast{float32}}[]<gpuarray> [id HH] ''   
 | |   |   | | | | |   | | | |   | | | |   | |     |   |Constant{2} [id DP]
 | |   |   | | | | |   | | | |   | | | |   | |     |InplaceGpuDimShuffle{x} [id HN] ''   
 | |   |   | | | | |   | | | |   | | | |   | |       |GpuSubtensor{int64} [id HO] ''   
 | |   |   | | | | |   | | | |   | | | |   | |         |GpuElemwise{Cast{float32}}[]<gpuarray> [id HH] ''   
 | |   |   | | | | |   | | | |   | | | |   | |         |Constant{3} [id DS]
 | |   |   | | | | |   | | | |   | | | |   | |InplaceGpuDimShuffle{x,0,x,x} [id HP] ''   
 | |   |   | | | | |   | | | |   | | | |   | | |gamma [id HQ]
 | |   |   | | | | |   | | | |   | | | |   | |InplaceGpuDimShuffle{x,0,x,x} [id HR] ''   
 | |   |   | | | | |   | | | |   | | | |   | | |GpuElemwise{inv,no_inplace} [id HS] ''   
 | |   |   | | | | |   | | | |   | | | |   | |   |GpuElemwise{Composite{sqrt((i0 + Cast{float16}((((i1 / i2) / i3) / i4))))}}[]<gpuarray> [id HT] ''   
 | |   |   | | | | |   | | | |   | | | |   | |     |GpuArrayConstant{[ 0.00010002]} [id DY]
 | |   |   | | | | |   | | | |   | | | |   | |     |GpuCAReduceCuda{add}{0, 2, 3} [id HU] ''   
 | |   |   | | | | |   | | | |   | | | |   | |     | |GpuElemwise{sqr,no_inplace} [id HV] ''   
 | |   |   | | | | |   | | | |   | | | |   | |     |   |GpuElemwise{Sub}[(0, 0)]<gpuarray> [id HW] ''   
 | |   |   | | | | |   | | | |   | | | |   | |     |     |GpuDnnConv{algo='small', inplace=True} [id BO] ''   
 | |   |   | | | | |   | | | |   | | | |   | |     |     |GpuElemwise{Composite{Cast{float16}((((i0 / i1) / i2) / i3))}}[]<gpuarray> [id HX] ''   
 | |   |   | | | | |   | | | |   | | | |   | |     |       |InplaceGpuDimShuffle{x,0,x,x} [id HY] ''   
 | |   |   | | | | |   | | | |   | | | |   | |     |       | |GpuCAReduceCuda{add}{0, 2, 3} [id HE] ''   
 | |   |   | | | | |   | | | |   | | | |   | |     |       |InplaceGpuDimShuffle{x,x,x,x} [id HZ] ''   
 | |   |   | | | | |   | | | |   | | | |   | |     |       | |GpuSubtensor{int64} [id HG] ''   
 | |   |   | | | | |   | | | |   | | | |   | |     |       |InplaceGpuDimShuffle{x,x,x,x} [id IA] ''   
 | |   |   | | | | |   | | | |   | | | |   | |     |       | |GpuSubtensor{int64} [id HM] ''   
 | |   |   | | | | |   | | | |   | | | |   | |     |       |InplaceGpuDimShuffle{x,x,x,x} [id IB] ''   
 | |   |   | | | | |   | | | |   | | | |   | |     |         |GpuSubtensor{int64} [id HO] ''   
 | |   |   | | | | |   | | | |   | | | |   | |     |InplaceGpuDimShuffle{x} [id HF] ''   
 | |   |   | | | | |   | | | |   | | | |   | |     |InplaceGpuDimShuffle{x} [id HL] ''   
 | |   |   | | | | |   | | | |   | | | |   | |     |InplaceGpuDimShuffle{x} [id HN] ''   
 | |   |   | | | | |   | | | |   | | | |   | |InplaceGpuDimShuffle{x,0,x,x} [id IC] ''   
 | |   |   | | | | |   | | | |   | | | |   |   |beta [id ID]
 | |   |   | | | | |   | | | |   | | | |   |GpuArrayConstant{[[[[ 0.49511719]]]]} [id EJ]
 | |   |   | | | | |   | | | |   | | | |GpuContiguous [id IE] ''   
 | |   |   | | | | |   | | | |   | | | | |W [id IF]
 | |   |   | | | | |   | | | |   | | | |GpuAllocEmpty{dtype='float16', context_name=None} [id IG] ''   
 | |   |   | | | | |   | | | |   | | | | |Assert{msg='The convolution would produce an invalid shape (dim[0] < 0).'} [id IH] ''   
 | |   |   | | | | |   | | | |   | | | | | |Shape_i{0} [id II] ''   
 | |   |   | | | | |   | | | |   | | | | | | |GpuContiguous [id BK] ''   
 | |   |   | | | | |   | | | |   | | | | | |Elemwise{ge,no_inplace} [id IJ] ''   
 | |   |   | | | | |   | | | |   | | | | |   |Shape_i{0} [id II] ''   
 | |   |   | | | | |   | | | |   | | | | |   |TensorConstant{0} [id CI]
 | |   |   | | | | |   | | | |   | | | | |Assert{msg='The convolution would produce an invalid shape (dim[1] < 0).'} [id IK] ''   
 | |   |   | | | | |   | | | |   | | | | | |Shape_i{0} [id IL] ''   
 | |   |   | | | | |   | | | |   | | | | | | |GpuContiguous [id IE] ''   
 | |   |   | | | | |   | | | |   | | | | | |Elemwise{ge,no_inplace} [id IM] ''   
 | |   |   | | | | |   | | | |   | | | | |   |Shape_i{0} [id IL] ''   
 | |   |   | | | | |   | | | |   | | | | |   |TensorConstant{0} [id CI]
 | |   |   | | | | |   | | | |   | | | | |Assert{msg='The convolution would produce an invalid shape (dim[2] <= 0).'} [id IN] ''   
 | |   |   | | | | |   | | | |   | | | | | |Elemwise{Composite{((((i0 + (i1 * (Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5) // i6))) - Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5)) // i7) + i8)}} [id IO] ''   
 | |   |   | | | | |   | | | |   | | | | | | |Shape_i{2} [id IP] ''   
 | |   |   | | | | |   | | | |   | | | | | | | |GpuContiguous [id BK] ''   
 | |   |   | | | | |   | | | |   | | | | | | |TensorConstant{2} [id CP]
 | |   |   | | | | |   | | | |   | | | | | | |Shape_i{2} [id IQ] ''   
 | |   |   | | | | |   | | | |   | | | | | | | |GpuContiguous [id IE] ''   
 | |   |   | | | | |   | | | |   | | | | | | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | | | | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | | | | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | | | | |TensorConstant{2} [id CP]
 | |   |   | | | | |   | | | |   | | | | | | |TensorConstant{2} [id CP]
 | |   |   | | | | |   | | | |   | | | | | | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | | | |Elemwise{gt,no_inplace} [id IR] ''   
 | |   |   | | | | |   | | | |   | | | | |   |Elemwise{Composite{((((i0 + (i1 * (Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5) // i6))) - Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5)) // i7) + i8)}} [id IO] ''   
 | |   |   | | | | |   | | | |   | | | | |   |TensorConstant{0} [id CI]
 | |   |   | | | | |   | | | |   | | | | |Assert{msg='The convolution would produce an invalid shape (dim[3] <= 0).'} [id IS] ''   
 | |   |   | | | | |   | | | |   | | | |   |Elemwise{Composite{((((i0 + (i1 * (Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5) // i6))) - Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5)) // i7) + i8)}} [id IT] ''   
 | |   |   | | | | |   | | | |   | | | |   | |Shape_i{3} [id IU] ''   
 | |   |   | | | | |   | | | |   | | | |   | | |GpuContiguous [id BK] ''   
 | |   |   | | | | |   | | | |   | | | |   | |TensorConstant{2} [id CP]
 | |   |   | | | | |   | | | |   | | | |   | |Shape_i{3} [id IV] ''   
 | |   |   | | | | |   | | | |   | | | |   | | |GpuContiguous [id IE] ''   
 | |   |   | | | | |   | | | |   | | | |   | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | |   | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | |   | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | |   | |TensorConstant{2} [id CP]
 | |   |   | | | | |   | | | |   | | | |   | |TensorConstant{2} [id CP]
 | |   |   | | | | |   | | | |   | | | |   | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | | | |   |Elemwise{gt,no_inplace} [id IW] ''   
 | |   |   | | | | |   | | | |   | | | |     |Elemwise{Composite{((((i0 + (i1 * (Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5) // i6))) - Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5)) // i7) + i8)}} [id IT] ''   
 | |   |   | | | | |   | | | |   | | | |     |TensorConstant{0} [id CI]
 | |   |   | | | | |   | | | |   | | | |GpuDnnConvDesc{border_mode='half', subsample=(2, 2), dilation=(1, 1), conv_mode='conv', precision='float32'} [id IX] ''   
 | |   |   | | | | |   | | | |   | | | | |Shape [id IY] ''   
 | |   |   | | | | |   | | | |   | | | |   |GpuContiguous [id IE] ''   
 | |   |   | | | | |   | | | |   | | | |Constant{1.0} [id DA]
 | |   |   | | | | |   | | | |   | | | |Constant{0.0} [id DB]
 | |   |   | | | | |   | | | |   | | |InplaceGpuDimShuffle{x,0,x,x} [id IZ] ''   
 | |   |   | | | | |   | | | |   | |   |GpuElemwise{Composite{Cast{float16}((((i0 / i1) / i2) / i3))}}[]<gpuarray> [id JA] ''   
 | |   |   | | | | |   | | | |   | |     |GpuCAReduceCuda{add}{0, 2, 3} [id JB] ''   
 | |   |   | | | | |   | | | |   | |     | |GpuDnnConv{algo='small', inplace=True} [id BJ] ''   
 | |   |   | | | | |   | | | |   | |     |InplaceGpuDimShuffle{x} [id JC] ''   
 | |   |   | | | | |   | | | |   | |     | |GpuSubtensor{int64} [id JD] ''   
 | |   |   | | | | |   | | | |   | |     |   |GpuElemwise{Cast{float32}}[]<gpuarray> [id JE] ''   
 | |   |   | | | | |   | | | |   | |     |   | |GpuFromHost<None> [id JF] ''   
 | |   |   | | | | |   | | | |   | |     |   |   |MakeVector{dtype='int64'} [id JG] ''   
 | |   |   | | | | |   | | | |   | |     |   |     |Shape_i{0} [id DK] ''   
 | |   |   | | | | |   | | | |   | |     |   |     |TensorConstant{64} [id HK]
 | |   |   | | | | |   | | | |   | |     |   |     |TensorConstant{8} [id JH]
 | |   |   | | | | |   | | | |   | |     |   |     |TensorConstant{8} [id JH]
 | |   |   | | | | |   | | | |   | |     |   |Constant{0} [id DM]
 | |   |   | | | | |   | | | |   | |     |InplaceGpuDimShuffle{x} [id JI] ''   
 | |   |   | | | | |   | | | |   | |     | |GpuSubtensor{int64} [id JJ] ''   
 | |   |   | | | | |   | | | |   | |     |   |GpuElemwise{Cast{float32}}[]<gpuarray> [id JE] ''   
 | |   |   | | | | |   | | | |   | |     |   |Constant{2} [id DP]
 | |   |   | | | | |   | | | |   | |     |InplaceGpuDimShuffle{x} [id JK] ''   
 | |   |   | | | | |   | | | |   | |       |GpuSubtensor{int64} [id JL] ''   
 | |   |   | | | | |   | | | |   | |         |GpuElemwise{Cast{float32}}[]<gpuarray> [id JE] ''   
 | |   |   | | | | |   | | | |   | |         |Constant{3} [id DS]
 | |   |   | | | | |   | | | |   | |InplaceGpuDimShuffle{x,0,x,x} [id JM] ''   
 | |   |   | | | | |   | | | |   | | |gamma [id JN]
 | |   |   | | | | |   | | | |   | |InplaceGpuDimShuffle{x,0,x,x} [id JO] ''   
 | |   |   | | | | |   | | | |   | | |GpuElemwise{inv,no_inplace} [id JP] ''   
 | |   |   | | | | |   | | | |   | |   |GpuElemwise{Composite{sqrt((i0 + Cast{float16}((((i1 / i2) / i3) / i4))))}}[]<gpuarray> [id JQ] ''   
 | |   |   | | | | |   | | | |   | |     |GpuArrayConstant{[ 0.00010002]} [id DY]
 | |   |   | | | | |   | | | |   | |     |GpuCAReduceCuda{add}{0, 2, 3} [id JR] ''   
 | |   |   | | | | |   | | | |   | |     | |GpuElemwise{sqr,no_inplace} [id JS] ''   
 | |   |   | | | | |   | | | |   | |     |   |GpuElemwise{sub,no_inplace} [id JT] ''   
 | |   |   | | | | |   | | | |   | |     |     |GpuDnnConv{algo='small', inplace=True} [id BJ] ''   
 | |   |   | | | | |   | | | |   | |     |     |GpuElemwise{Composite{Cast{float16}((((i0 / i1) / i2) / i3))}}[]<gpuarray> [id JU] ''   
 | |   |   | | | | |   | | | |   | |     |       |InplaceGpuDimShuffle{x,0,x,x} [id JV] ''   
 | |   |   | | | | |   | | | |   | |     |       | |GpuCAReduceCuda{add}{0, 2, 3} [id JB] ''   
 | |   |   | | | | |   | | | |   | |     |       |InplaceGpuDimShuffle{x,x,x,x} [id JW] ''   
 | |   |   | | | | |   | | | |   | |     |       | |GpuSubtensor{int64} [id JD] ''   
 | |   |   | | | | |   | | | |   | |     |       |InplaceGpuDimShuffle{x,x,x,x} [id JX] ''   
 | |   |   | | | | |   | | | |   | |     |       | |GpuSubtensor{int64} [id JJ] ''   
 | |   |   | | | | |   | | | |   | |     |       |InplaceGpuDimShuffle{x,x,x,x} [id JY] ''   
 | |   |   | | | | |   | | | |   | |     |         |GpuSubtensor{int64} [id JL] ''   
 | |   |   | | | | |   | | | |   | |     |InplaceGpuDimShuffle{x} [id JC] ''   
 | |   |   | | | | |   | | | |   | |     |InplaceGpuDimShuffle{x} [id JI] ''   
 | |   |   | | | | |   | | | |   | |     |InplaceGpuDimShuffle{x} [id JK] ''   
 | |   |   | | | | |   | | | |   | |InplaceGpuDimShuffle{x,0,x,x} [id JZ] ''   
 | |   |   | | | | |   | | | |   |   |beta [id KA]
 | |   |   | | | | |   | | | |   |GpuArrayConstant{[[[[ 0.49511719]]]]} [id EJ]
 | |   |   | | | | |   | | | |GpuContiguous [id KB] ''   
 | |   |   | | | | |   | | | | |W [id KC]
 | |   |   | | | | |   | | | |GpuAllocEmpty{dtype='float16', context_name=None} [id KD] ''   
 | |   |   | | | | |   | | | | |Assert{msg='The convolution would produce an invalid shape (dim[0] < 0).'} [id KE] ''   
 | |   |   | | | | |   | | | | | |Shape_i{0} [id KF] ''   
 | |   |   | | | | |   | | | | | | |GpuContiguous [id BF] ''   
 | |   |   | | | | |   | | | | | |Elemwise{ge,no_inplace} [id KG] ''   
 | |   |   | | | | |   | | | | |   |Shape_i{0} [id KF] ''   
 | |   |   | | | | |   | | | | |   |TensorConstant{0} [id CI]
 | |   |   | | | | |   | | | | |Assert{msg='The convolution would produce an invalid shape (dim[1] < 0).'} [id KH] ''   
 | |   |   | | | | |   | | | | | |Shape_i{0} [id KI] ''   
 | |   |   | | | | |   | | | | | | |GpuContiguous [id KB] ''   
 | |   |   | | | | |   | | | | | |Elemwise{ge,no_inplace} [id KJ] ''   
 | |   |   | | | | |   | | | | |   |Shape_i{0} [id KI] ''   
 | |   |   | | | | |   | | | | |   |TensorConstant{0} [id CI]
 | |   |   | | | | |   | | | | |Assert{msg='The convolution would produce an invalid shape (dim[2] <= 0).'} [id KK] ''   
 | |   |   | | | | |   | | | | | |Elemwise{Composite{(((i0 + (i1 * (Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5) // i6))) - Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5)) + i7)}} [id KL] ''   
 | |   |   | | | | |   | | | | | | |Shape_i{2} [id KM] ''   
 | |   |   | | | | |   | | | | | | | |GpuContiguous [id BF] ''   
 | |   |   | | | | |   | | | | | | |TensorConstant{2} [id CP]
 | |   |   | | | | |   | | | | | | |Shape_i{2} [id KN] ''   
 | |   |   | | | | |   | | | | | | | |GpuContiguous [id KB] ''   
 | |   |   | | | | |   | | | | | | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | | | | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | | | | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | | | | |TensorConstant{2} [id CP]
 | |   |   | | | | |   | | | | | | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | | | |Elemwise{gt,no_inplace} [id KO] ''   
 | |   |   | | | | |   | | | | |   |Elemwise{Composite{(((i0 + (i1 * (Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5) // i6))) - Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5)) + i7)}} [id KL] ''   
 | |   |   | | | | |   | | | | |   |TensorConstant{0} [id CI]
 | |   |   | | | | |   | | | | |Assert{msg='The convolution would produce an invalid shape (dim[3] <= 0).'} [id KP] ''   
 | |   |   | | | | |   | | | |   |Elemwise{Composite{(((i0 + (i1 * (Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5) // i6))) - Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5)) + i7)}} [id KQ] ''   
 | |   |   | | | | |   | | | |   | |Shape_i{3} [id KR] ''   
 | |   |   | | | | |   | | | |   | | |GpuContiguous [id BF] ''   
 | |   |   | | | | |   | | | |   | |TensorConstant{2} [id CP]
 | |   |   | | | | |   | | | |   | |Shape_i{3} [id KS] ''   
 | |   |   | | | | |   | | | |   | | |GpuContiguous [id KB] ''   
 | |   |   | | | | |   | | | |   | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   | |TensorConstant{2} [id CP]
 | |   |   | | | | |   | | | |   | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | | | |   |Elemwise{gt,no_inplace} [id KT] ''   
 | |   |   | | | | |   | | | |     |Elemwise{Composite{(((i0 + (i1 * (Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5) // i6))) - Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5)) + i7)}} [id KQ] ''   
 | |   |   | | | | |   | | | |     |TensorConstant{0} [id CI]
 | |   |   | | | | |   | | | |GpuDnnConvDesc{border_mode='half', subsample=(1, 1), dilation=(1, 1), conv_mode='conv', precision='float32'} [id KU] ''   
 | |   |   | | | | |   | | | | |Shape [id KV] ''   
 | |   |   | | | | |   | | | |   |GpuContiguous [id KB] ''   
 | |   |   | | | | |   | | | |Constant{1.0} [id DA]
 | |   |   | | | | |   | | | |Constant{0.0} [id DB]
 | |   |   | | | | |   | | |InplaceGpuDimShuffle{x,0,x,x} [id KW] ''   
 | |   |   | | | | |   | |   |GpuElemwise{Composite{Cast{float16}((((i0 / i1) / i2) / i3))}}[]<gpuarray> [id KX] ''   
 | |   |   | | | | |   | |     |GpuCAReduceCuda{add}{0, 2, 3} [id KY] ''   
 | |   |   | | | | |   | |     | |GpuDnnConv{algo='small', inplace=True} [id BE] ''   
 | |   |   | | | | |   | |     |InplaceGpuDimShuffle{x} [id KZ] ''   
 | |   |   | | | | |   | |     | |GpuSubtensor{int64} [id LA] ''   
 | |   |   | | | | |   | |     |   |GpuElemwise{Cast{float32}}[]<gpuarray> [id LB] ''   
 | |   |   | | | | |   | |     |   | |GpuFromHost<None> [id LC] ''   
 | |   |   | | | | |   | |     |   |   |MakeVector{dtype='int64'} [id LD] ''   
 | |   |   | | | | |   | |     |   |     |Shape_i{0} [id DK] ''   
 | |   |   | | | | |   | |     |   |     |TensorConstant{128} [id LE]
 | |   |   | | | | |   | |     |   |     |TensorConstant{8} [id JH]
 | |   |   | | | | |   | |     |   |     |TensorConstant{8} [id JH]
 | |   |   | | | | |   | |     |   |Constant{0} [id DM]
 | |   |   | | | | |   | |     |InplaceGpuDimShuffle{x} [id LF] ''   
 | |   |   | | | | |   | |     | |GpuSubtensor{int64} [id LG] ''   
 | |   |   | | | | |   | |     |   |GpuElemwise{Cast{float32}}[]<gpuarray> [id LB] ''   
 | |   |   | | | | |   | |     |   |Constant{2} [id DP]
 | |   |   | | | | |   | |     |InplaceGpuDimShuffle{x} [id LH] ''   
 | |   |   | | | | |   | |       |GpuSubtensor{int64} [id LI] ''   
 | |   |   | | | | |   | |         |GpuElemwise{Cast{float32}}[]<gpuarray> [id LB] ''   
 | |   |   | | | | |   | |         |Constant{3} [id DS]
 | |   |   | | | | |   | |InplaceGpuDimShuffle{x,0,x,x} [id LJ] ''   
 | |   |   | | | | |   | | |gamma [id LK]
 | |   |   | | | | |   | |InplaceGpuDimShuffle{x,0,x,x} [id LL] ''   
 | |   |   | | | | |   | | |GpuElemwise{inv,no_inplace} [id LM] ''   
 | |   |   | | | | |   | |   |GpuElemwise{Composite{sqrt((i0 + Cast{float16}((((i1 / i2) / i3) / i4))))}}[]<gpuarray> [id LN] ''   
 | |   |   | | | | |   | |     |GpuArrayConstant{[ 0.00010002]} [id DY]
 | |   |   | | | | |   | |     |GpuCAReduceCuda{add}{0, 2, 3} [id LO] ''   
 | |   |   | | | | |   | |     | |GpuElemwise{sqr,no_inplace} [id LP] ''   
 | |   |   | | | | |   | |     |   |GpuElemwise{sub,no_inplace} [id LQ] ''   
 | |   |   | | | | |   | |     |     |GpuDnnConv{algo='small', inplace=True} [id BE] ''   
 | |   |   | | | | |   | |     |     |GpuElemwise{Composite{Cast{float16}((((i0 / i1) / i2) / i3))}}[]<gpuarray> [id LR] ''   
 | |   |   | | | | |   | |     |       |InplaceGpuDimShuffle{x,0,x,x} [id LS] ''   
 | |   |   | | | | |   | |     |       | |GpuCAReduceCuda{add}{0, 2, 3} [id KY] ''   
 | |   |   | | | | |   | |     |       |InplaceGpuDimShuffle{x,x,x,x} [id LT] ''   
 | |   |   | | | | |   | |     |       | |GpuSubtensor{int64} [id LA] ''   
 | |   |   | | | | |   | |     |       |InplaceGpuDimShuffle{x,x,x,x} [id LU] ''   
 | |   |   | | | | |   | |     |       | |GpuSubtensor{int64} [id LG] ''   
 | |   |   | | | | |   | |     |       |InplaceGpuDimShuffle{x,x,x,x} [id LV] ''   
 | |   |   | | | | |   | |     |         |GpuSubtensor{int64} [id LI] ''   
 | |   |   | | | | |   | |     |InplaceGpuDimShuffle{x} [id KZ] ''   
 | |   |   | | | | |   | |     |InplaceGpuDimShuffle{x} [id LF] ''   
 | |   |   | | | | |   | |     |InplaceGpuDimShuffle{x} [id LH] ''   
 | |   |   | | | | |   | |InplaceGpuDimShuffle{x,0,x,x} [id LW] ''   
 | |   |   | | | | |   |   |beta [id LX]
 | |   |   | | | | |   |GpuArrayConstant{[[[[ 0.49511719]]]]} [id EJ]
 | |   |   | | | | |GpuContiguous [id LY] ''   
 | |   |   | | | | | |W [id LZ]
 | |   |   | | | | |GpuAllocEmpty{dtype='float16', context_name=None} [id MA] ''   
 | |   |   | | | | | |Assert{msg='The convolution would produce an invalid shape (dim[0] < 0).'} [id MB] ''   
 | |   |   | | | | | | |Shape_i{0} [id MC] ''   
 | |   |   | | | | | | | |GpuContiguous [id BA] ''   
 | |   |   | | | | | | |Elemwise{ge,no_inplace} [id MD] ''   
 | |   |   | | | | | |   |Shape_i{0} [id MC] ''   
 | |   |   | | | | | |   |TensorConstant{0} [id CI]
 | |   |   | | | | | |Assert{msg='The convolution would produce an invalid shape (dim[1] < 0).'} [id ME] ''   
 | |   |   | | | | | | |Shape_i{0} [id MF] ''   
 | |   |   | | | | | | | |GpuContiguous [id LY] ''   
 | |   |   | | | | | | |Elemwise{ge,no_inplace} [id MG] ''   
 | |   |   | | | | | |   |Shape_i{0} [id MF] ''   
 | |   |   | | | | | |   |TensorConstant{0} [id CI]
 | |   |   | | | | | |Assert{msg='The convolution would produce an invalid shape (dim[2] <= 0).'} [id MH] ''   
 | |   |   | | | | | | |Elemwise{Composite{((((i0 + (i1 * (Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5) // i6))) - Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5)) // i7) + i8)}} [id MI] ''   
 | |   |   | | | | | | | |Shape_i{2} [id MJ] ''   
 | |   |   | | | | | | | | |GpuContiguous [id BA] ''   
 | |   |   | | | | | | | |TensorConstant{2} [id CP]
 | |   |   | | | | | | | |Shape_i{2} [id MK] ''   
 | |   |   | | | | | | | | |GpuContiguous [id LY] ''   
 | |   |   | | | | | | | |TensorConstant{1} [id CR]
 | |   |   | | | | | | | |TensorConstant{1} [id CR]
 | |   |   | | | | | | | |TensorConstant{1} [id CR]
 | |   |   | | | | | | | |TensorConstant{2} [id CP]
 | |   |   | | | | | | | |TensorConstant{2} [id CP]
 | |   |   | | | | | | | |TensorConstant{1} [id CR]
 | |   |   | | | | | | |Elemwise{gt,no_inplace} [id ML] ''   
 | |   |   | | | | | |   |Elemwise{Composite{((((i0 + (i1 * (Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5) // i6))) - Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5)) // i7) + i8)}} [id MI] ''   
 | |   |   | | | | | |   |TensorConstant{0} [id CI]
 | |   |   | | | | | |Assert{msg='The convolution would produce an invalid shape (dim[3] <= 0).'} [id MM] ''   
 | |   |   | | | | |   |Elemwise{Composite{((((i0 + (i1 * (Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5) // i6))) - Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5)) // i7) + i8)}} [id MN] ''   
 | |   |   | | | | |   | |Shape_i{3} [id MO] ''   
 | |   |   | | | | |   | | |GpuContiguous [id BA] ''   
 | |   |   | | | | |   | |TensorConstant{2} [id CP]
 | |   |   | | | | |   | |Shape_i{3} [id MP] ''   
 | |   |   | | | | |   | | |GpuContiguous [id LY] ''   
 | |   |   | | | | |   | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | |TensorConstant{1} [id CR]
 | |   |   | | | | |   | |TensorConstant{2} [id CP]
 | |   |   | | | | |   | |TensorConstant{2} [id CP]
 | |   |   | | | | |   | |TensorConstant{1} [id CR]
 | |   |   | | | | |   |Elemwise{gt,no_inplace} [id MQ] ''   
 | |   |   | | | | |     |Elemwise{Composite{((((i0 + (i1 * (Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5) // i6))) - Composite{(((i0 - i1) * i2) + i3)}(i2, i3, i4, i5)) // i7) + i8)}} [id MN] ''   
 | |   |   | | | | |     |TensorConstant{0} [id CI]
 | |   |   | | | | |GpuDnnConvDesc{border_mode='half', subsample=(2, 2), dilation=(1, 1), conv_mode='conv', precision='float32'} [id MR] ''   
 | |   |   | | | | | |Shape [id MS] ''   
 | |   |   | | | | |   |GpuContiguous [id LY] ''   
 | |   |   | | | | |Constant{1.0} [id DA]
 | |   |   | | | | |Constant{0.0} [id DB]
 | |   |   | | | |InplaceGpuDimShuffle{x,0,x,x} [id MT] ''   
 | |   |   | | |   |GpuElemwise{Composite{Cast{float16}((((i0 / i1) / i2) / i3))}}[]<gpuarray> [id MU] ''   
 | |   |   | | |     |GpuCAReduceCuda{add}{0, 2, 3} [id MV] ''   
 | |   |   | | |     | |GpuDnnConv{algo='small', inplace=True} [id Z] ''   
 | |   |   | | |     |InplaceGpuDimShuffle{x} [id MW] ''   
 | |   |   | | |     | |GpuSubtensor{int64} [id MX] ''   
 | |   |   | | |     |   |GpuElemwise{Cast{float32}}[]<gpuarray> [id MY] ''   
 | |   |   | | |     |   | |GpuFromHost<None> [id MZ] ''   
 | |   |   | | |     |   |   |MakeVector{dtype='int64'} [id NA] ''   
 | |   |   | | |     |   |     |Shape_i{0} [id DK] ''   
 | |   |   | | |     |   |     |TensorConstant{128} [id LE]
 | |   |   | | |     |   |     |TensorConstant{4} [id NB]
 | |   |   | | |     |   |     |TensorConstant{4} [id NB]
 | |   |   | | |     |   |Constant{0} [id DM]
 | |   |   | | |     |InplaceGpuDimShuffle{x} [id NC] ''   
 | |   |   | | |     | |GpuSubtensor{int64} [id ND] ''   
 | |   |   | | |     |   |GpuElemwise{Cast{float32}}[]<gpuarray> [id MY] ''   
 | |   |   | | |     |   |Constant{2} [id DP]
 | |   |   | | |     |InplaceGpuDimShuffle{x} [id NE] ''   
 | |   |   | | |       |GpuSubtensor{int64} [id NF] ''   
 | |   |   | | |         |GpuElemwise{Cast{float32}}[]<gpuarray> [id MY] ''   
 | |   |   | | |         |Constant{3} [id DS]
 | |   |   | | |InplaceGpuDimShuffle{x,0,x,x} [id NG] ''   
 | |   |   | | | |gamma [id NH]
 | |   |   | | |InplaceGpuDimShuffle{x,0,x,x} [id NI] ''   
 | |   |   | | | |GpuElemwise{inv,no_inplace} [id NJ] ''   
 | |   |   | | |   |GpuElemwise{Composite{sqrt((i0 + Cast{float16}((((i1 / i2) / i3) / i4))))}}[]<gpuarray> [id NK] ''   
 | |   |   | | |     |GpuArrayConstant{[ 0.00010002]} [id DY]
 | |   |   | | |     |GpuCAReduceCuda{add}{0, 2, 3} [id NL] ''   
 | |   |   | | |     | |GpuElemwise{sqr,no_inplace} [id NM] ''   
 | |   |   | | |     |   |GpuElemwise{sub,no_inplace} [id NN] ''   
 | |   |   | | |     |     |GpuDnnConv{algo='small', inplace=True} [id Z] ''   
 | |   |   | | |     |     |GpuElemwise{Composite{Cast{float16}((((i0 / i1) / i2) / i3))}}[]<gpuarray> [id NO] ''   
 | |   |   | | |     |       |InplaceGpuDimShuffle{x,0,x,x} [id NP] ''   
 | |   |   | | |     |       | |GpuCAReduceCuda{add}{0, 2, 3} [id MV] ''   
 | |   |   | | |     |       |InplaceGpuDimShuffle{x,x,x,x} [id NQ] ''   
 | |   |   | | |     |       | |GpuSubtensor{int64} [id MX] ''   
 | |   |   | | |     |       |InplaceGpuDimShuffle{x,x,x,x} [id NR] ''   
 | |   |   | | |     |       | |GpuSubtensor{int64} [id ND] ''   
 | |   |   | | |     |       |InplaceGpuDimShuffle{x,x,x,x} [id NS] ''   
 | |   |   | | |     |         |GpuSubtensor{int64} [id NF] ''   
 | |   |   | | |     |InplaceGpuDimShuffle{x} [id MW] ''   
 | |   |   | | |     |InplaceGpuDimShuffle{x} [id NC] ''   
 | |   |   | | |     |InplaceGpuDimShuffle{x} [id NE] ''   
 | |   |   | | |InplaceGpuDimShuffle{x,0,x,x} [id NT] ''   
 | |   |   | |   |beta [id NU]
 | |   |   | |GpuArrayConstant{[[[[ 0.49511719]]]]} [id EJ]
 | |   |   |MakeVector{dtype='int64'} [id NV] ''   
 | |   |     |Shape_i{0} [id DK] ''   
 | |   |     |TensorConstant{128} [id LE]
 | |   |     |TensorConstant{-1} [id NW]
 | |   |InplaceGpuDimShuffle{x,x} [id NX] ''   
 | |     |GpuElemwise{Cast{float32}}[]<gpuarray> [id NY] ''   
 | |       |GpuFromHost<None> [id NZ] ''   
 | |         |Elemwise{Composite{((i0 * i1) // maximum((i2 * i1), i3))}} [id OA] ''   
 | |           |TensorConstant{2048} [id OB]
 | |           |Shape_i{0} [id DK] ''   
 | |           |TensorConstant{128} [id LE]
 | |           |TensorConstant{1} [id OC]
 | |GpuCrossentropySoftmax1HotWithBiasDx [id OD] ''   
 | | |GpuElemwise{Composite{Cast{float16}((i0 / i1))}}[]<gpuarray> [id OE] ''   
 | | | |GpuArrayConstant{[ 2048.]} [id OF]
 | | | |InplaceGpuDimShuffle{x} [id OG] ''   
 | | |   |GpuElemwise{Cast{float32}}[]<gpuarray> [id OH] ''   
 | | |     |GpuFromHost<None> [id OI] ''   
 | | |       |Shape_i{0} [id DK] ''   
 | | |GpuCrossentropySoftmaxArgmax1HotWithBias.1 [id OJ] ''   
 | | | |GpuDot22 [id OK] ''   
 | | | | |GpuElemwise{Composite{Cast{float16}((i0 / i1))}}[]<gpuarray> [id S] ''   
 | | | | |W [id B]
 | | | |b [id OL]
 | | | |GpuFromHost<None> [id OM] ''   
 | | |   |<TensorType(int32, vector)> [id ON]
 | | |GpuFromHost<None> [id OM] ''   
 | |TensorConstant{1.0} [id OO]
 |GpuElemwise{Composite{((i0 * i1) + (i2 * sqr(i3)))}}[(0, 1)]<gpuarray> [id OP] ''   
   |GpuArrayConstant{[[ 0.99902344]]} [id OQ]
   |<GpuArrayType<None>(float16, matrix)> [id OR]
   |GpuArrayConstant{[[ 0.00097656]]} [id OS]
   |GpuDot22 [id OT] ''   
     |InplaceGpuDimShuffle{1,0} [id R] ''   
     |GpuCrossentropySoftmax1HotWithBiasDx [id OD] ''   



Apply node that caused the error: GpuElemwise{Composite{(i0 - ((i1 * i2) / sqrt(i3)))}}[(0, 0)]<gpuarray>(W, InplaceGpuDimShuffle{x,x}.0, GpuGemm{inplace=True}.0, GpuElemwise{Composite{((i0 * i1) + (i2 * sqr(i3)))}}[(0, 1)]<gpuarray>.0)
Toposort index: 421
Inputs types: [GpuArrayType<None>(float16, matrix), GpuArrayType<None>(float16, (True, True)), GpuArrayType<None>(float16, matrix), GpuArrayType<None>(float16, matrix)]
Inputs shapes: [(128, 10), (1, 1), (128, 10), (128, 10)]
Inputs strides: [(20, 2), (2, 2), (20, 2), (20, 2)]
Inputs values: ['not shown', gpuarray.array([[ 0.00156116]], dtype=float16), 'not shown', 'not shown']
Outputs clients: [['output']]

HINT: Re-running with most Theano optimization disabled could give you a back-trace of when this node was created. This can be done with by setting the Theano flag 'optimizer=fast_compile'. If that does not work, Theano optimizations can be disabled with 'optimizer=None'.
HINT: Use the Theano flag 'exception_verbosity=high' for a debugprint and storage map footprint of this apply node.