Python stopped working when execute 'self.solver.step(1)' code in 'train.py' file.

[cobbwho]

I completed the caffe compilation in cpu-only mode on windows without cuda. Because I don't have linux and no cuda. When I train with the voc_2007 data set via train_net.py, Some errors have been made.When the program runs to the position shown in the figure below:
...
I0423 net.cpp:774] Copying source layer fc7 I0423 net.cpp:774] Copying source layer relu7 I0423 net.cpp:774] Copying source layer drop7 I0423 net.cpp:771] Ignoring source layer fc8 I0423 net.cpp:771] Ignoring source layer loss
Solving...
Then, the dialog box indicates that python.exe has stopped working and no error message. I don't kown what‘s the wrong and the program crushed.
I tried to find the error code in the train.py file and I find the program stoped in code self.solver.step(1) in 92 rows and 12 columns. It seems like that error occurred in the underlying cpp file to check something about gpu , but I did not find.
Anyone got any ideas? Any word will be appreciate.

[cobbwho]

I have made some progress when I modify solver.prototxt and set fileld 'debug_info' to true. Some new messages occured.
...like before
Solving...
........Omitted
I0424 net.cpp:630] [Forward] Layer cls_score, param blob 0 data: 0.00795975
I0424 net.cpp:630] [Forward] Layer cls_score, param blob 1 data: 0
I0424 net.cpp:618] [Forward] Layer bbox_pred, top blob bbox_pred data: 0.0848037
I0424 net.cpp:630] [Forward] Layer bbox_pred, param blob 0 data: 0.000797856
I0424 net.cpp:630] [Forward] Layer bbox_pred, param blob 1 data: 0
I0424 net.cpp:618] [Forward] Layer loss_cls, top blob loss_cls data: 4.00548
and then, python.exe still stop working.
Fortunately, I actually modify file smooth_L1_loss_layer.cpp to support cpu mode about forward delivery and backward delivery. It will report an error not implemented yet before I modify the file. And now may be the error because smooth_L1_loss_layer.cpp

[cobbwho]

Unfortunately, the author rbgirshick already said they do not plan to support CPU-only training Cpu only #39.Although I have modified ‘smooth_L1_loss_layer.cpp’ and ‘roi_pooling_layer.cpp’ and add forward backward method, It's work. I don't have the time and the ability to modify the underlying code to make it support cpu_only.