rob-hen
rob-hen
@patrickvonplaten Was looking for this answer in the documentation. Would be great to have this more prominent in the main doc.
Hi @zhanghongyong123456, thank you for your interest in our code. The code has been released.
Hi, indeed our current code release can be run with 12 GB VRAM. We are working on reducing the memory requirement even further.
Our latest code can be tested with less than 7GB VRAM.
Once all images have been created, a video is generated. The fps for the resulting video is configured in this line: https://github.com/Picsart-AI-Research/Text2Video-Zero/blob/fd76734e06aadb1dee83d2b7a368b14bdaa35565/utils.py#L113 .
We provide the option to process a video in chunks. Using that you can process long videos while the VRAM usage is constant for each chunk.
I encountered the same issue. I was manually setting `p.requires_grad=False` for some parameters. When I was doing this within the `configure_optimizers` method, I got the same error message as @athn-nik....
At least for logging with MLflow to the local storage, I created a workaround by broadcasting the artifact folder and manually copying the data into that folder. But this will...
Thanks for your answer @awaelchli I am not using wandb but MLFlow. Since all workers use a shared file system, I could have each worker write into a common folder,...
So your answer implies that `on_validation_epoch_end ` is already synchronized across all ranks, is that correct?