GAN Manual optimization not working after 1.0.7

[import-antigravity]

🐛 Bug

I have a GAN model and as of 1.0.7 it no longer trains correctly. I've tried to troubleshoot the issue to the best of my ability but I have no idea what's causing the problem.

Please reproduce using the BoringModel

https://colab.research.google.com/gist/import-antigravity/0730243bb11b56031110fd6aa7d58971/the-boringmodel.ipynb

To Reproduce

See boringmodel above^^^

Expected behavior

Using the colab notebook switch between version 1.0.6 and 1.0.7 to see the bug. For 1.0.6 after training a few epochs it's clear the GAN is beginning to converge, for 1.0.7 it just makes noise.

Environment

• PyTorch Version (e.g., 1.0): 1.0.7
• OS (e.g., Linux): macOS, Linux
• How you installed PyTorch (conda, pip, source): conda
• Build command you used (if compiling from source): n/a
• Python version: 3.8
• CUDA/cuDNN version: N/A
• GPU models and configuration: N/A
• Any other relevant information: N/A

[tchaton]

Hey there,

Would you mind to try out master ?

Best,
T.C

[import-antigravity]

The error happens on every version up to and including the current stable release.

…

On Jan 17, 2021, at 3:49 AM, chaton ***@***.***> wrote:  Hey there, Would you mind to try out master ? Best, T.C — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.

[Borda]

Hello, I have run the attached example but I have not seen any error, mind pls clarify what is the expected behavior?

[import-antigravity]

Running with version 1.0.7 produces basically noise, as shown in the ‘imshow’ cell

[import-antigravity]

It’s not an error, the code runs but does not function correcty

[import-antigravity]

Hello, I have run the attached example but I have not seen any error, mind pls clarify what is the expected behavior?

Here is the notebook run with 1.0.6, you can see that after a couple of epochs it's clearly converging to a digit:

Here is the same notebook run with 1.0.7, there's no such convergence and it just produces noise:

[import-antigravity]

@Borda are you able to reproduce the above images in the notebook?

[Borda]

@Borda are you able to reproduce the above images in the notebook?

yes, I can reproduce and see the same diff you illustrated in #5545 (comment)

[tchaton]

Hey @import-antigravity,

I have bumped this to P0 Priority. We will look into it asap.

Best,
T.C