A problem with RAM. DSQN.

[timswim]

For faster response

| Reinforcement Learning | lucifer2859

Issue type

• Bug Report
• Feature Request
• Help wanted
• Other

SpikingJelly version

master

Description

Hello, I am working on the DSQN project, but I can't finish the training, because the RAM is gradually filling up (within a few thousand frames after the training has started). Could this be a bug in the code/compatibility issue or do I just need more memory (I have 32GB)?

Thanks in advance!

# Lines of code where memory is filled
loss_v = common.calc_loss_dqn(batch, net, tgt_net.target_model, gamma=params['gamma'], cuda=args.cuda, cuda_async=True)
loss_v.backward()