philgzl

Solved it for me too. Cloning the current master branch and merging this fix gets me the desired behavior. ```bash git clone https://github.com/Bismuth-Forge/bismuth.git cd bismuth git remote add benemorius https://github.com/benemorius/bismuth.git...

> Maybe worth noting, but the batch_size option set to anything but 1 is going to break on mps. Right. We could hard set `batch_size=1` with MPS and raise a...

Still haven't started working in applying the suggested changes, will do it soon

Python: 3.10.7 OS: Scientific Linux 7.9 (Nitrogen) It seems like what is causing the issue is setting `vae=AutoencoderKL.from_pretrained(f"stabilityai/sd-vae-ft-ema")` when initializing the `StableDiffusionWalkPipeline` [here](https://github.com/nateraw/stable-diffusion-videos/blob/1207867acd2abff78f0f932914627cec1805bb6d/examples/make_music_video.py#L10). This is not done in the code...

@mthrok Maybe the reason why this was implemented this way is because `torch.stft` discards trailing samples if they do not fit a frame. This is specially noticeable when using `center=False`...

Thanks!

@deependujha inheriting from `StreamingDataset` does indeed suppress the type check error in `StreamingDataLoader.__init__` , but as pointed out by @tchaton it does not make `StreamingDataLoader` provide all its functionalities to...

Will give it a try. Use case is dynamic acoustic scene simulation for speech enhancement or source separation. I have datasets of clean speech files, datasets of noise files, and...

I think this is because worker 4 is assigned 0 samples from dataset 1 (dataset length is 10, batch size is 4 and number of workers is 4). The reasoning...

I thought I would give an example describing what I would like to achieve. Maybe this can help the review or spark ideas to improve the PR. I have two...