Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Hello,
I wanted to report an issue that I encountered while training on the Vimeo90k dataset using PyTorch 1.6 or higher, and I believe it to be a bug.
The problem arises from the fact that the random seed is no longer fixed with the 'random.seed(seed)' code in PyTorch versions 1.6 and later. As a result, the training process does not work as expected.
To address this issue, I propose a simple fix by adding 'torch.manual_seed(seed)' to the code. By including this line, the random seed will be properly set, leading to consistent and reliable training results.
I have tested this solution with PyTorch 2.0 and CUDA 12.0, and I can confirm that the code works flawlessly after implementing this fix.
Best regards,
Jihoon Nam