[BUG] Training stucks for Action Recognition

[Deepanshu-beep]

Description

Training stops for Action Recognition on 1st epoch at about batch 30/40 of 860 while training on HMDB51. The cell keeps on running but still the training is like paused there. But code works fine for demo task

In which platform does it happen?

Linux 18.0.4 Ubuntu [GCC 8.4.0] , PyTorch 1.4.0 , Torch-vision 0.5.0, Tesla T4

How do we replicate the issue?

Expected behavior (i.e. solution)/

Other Comments