`lock.acquire()` and cannot exit training

[Bostoncake]

Hi! I tried to reproduce your work under the Dassl framework as stated in README. However, in some of the training cases, my program cannot exit normally after finishing the training, although the training results are already printed. I tried to Ctrl+C the program to manually exit the training, and my Python reports are as followings:

This happens only occasionally, which means, only around 1/3 training cases that this happens, in the remaining 2/3, the training exits normally. I wonder if you have any insights towards this problem.
Thanks!