Memory required explodes when the CapsLayer instance is called

[arshiyaaggarwal]

@cedrickchee The memory required by the network increases from 3 GB to about 25 GB after the finishing of the Primary caps layer and after Capslayer is called. I wonder why so. I'm not able to solve this issue. The batch size is (8,128,64,6) which is not too large. On checking through print statements it is the classes Capslayer and Agreement routing that are causing an issue. Please help me out.

[cedrickchee]

Hi @arshiyaaggarwal Thank you for reporting this. My plate is quite full at the moment. I will try my best to look into this tomorrow. I don't fully understand the problem based on your explanation. It's better if you can send me some screen shots when the problem occured.

Here are my questions:

1. Is it possible for you to provide more info?

• your training environment details: OS, GPU spec, CPU spec
• software versions: Python, PyTorch

2. Hyperparameters passed to the main training script

• what are the commands you run?

3. Logs

• stack traces
• log from debugging

Let me know if you are not clear.