Error detected in SoftmaxBackward

[siowan]

Hello, I tried to use the SOF dataset to run in main_func.py. But it returns the error:

[W python_anomaly_mode.cpp:60] Warning: Error detected in SoftmaxBackward. Traceback of forward call that caused the error:
File "main_func.py", line 206, in
model = train_eval_sahp(params)

Could you share some ideas to let me know how to fix it? Thank you very much!

[QiangAIResearcher]

I'm not sure why you met such a problem, but you can try to identify the line that produce bugs by
with autograd.detect_anomaly()

Then you can print out the variable values and gradients of the line at question.

[siowan]

Thank you for your response, the problem is solved when I try to decrease the default batch size.