Hello, I am on Windows 10, and my gpu is a PNY Nvidia GTX 1660 TI 6 Gb. I installed V-Diffusion like so:

• conda create --name v-diffusion python=3.8
• conda activate v-diffusion
• conda install pytorch torchvision torchaudio cudatoolkit=11.3 -c pytorch (as per Pytorch website instructions)
• pip install requests tqdm

The problem is that when I launch the cfg_sample.py or clip_sample.py command lines, the generated images are completely black, although the inference process seems to run nicely and without errors.

Things I've tried:

• installing previous pytorch version with conda install pytorch torchvision torchaudio cudatoolkit=10.2 -c pytorch
• removing V-Diffusion conda environment completely and recreating it anew
• uninstalling nvidia drivers and performing a new clean driver install (I tried both Nvidia Studio drivers and Nvidia Game Ready drivers)
• uninstalling and reinstalling Conda completely

But nothing helped... and at this point I don't know what else to try...

The only interesting piece of information I could gather is that for some reason this problem also happens with another text-to-image project called Big Sleep where similar to V-Diffusion the inference process appears to run correctly but the generated images are all black.

I think there must be some simple detail I'm overlooking... which it's making me go insane... 😵 Please let me know something if you think you can help! THANKS !

A weight of 1 will sample images that match the prompt roughly as well as images usually match prompts like that in the training set.

I can't wrap my head around this sentence. Could you please explain it with different wording? Thanks!

torch version: 1.8.1+cu111

python ./cfg_sample.py "the rise of consciousness":5 -n 4 -bs 4 --seed 0 Using device: cuda:0 Traceback (most recent call last): File "./cfg_sample.py", line 154, in main() File "./cfg_sample.py", line 148, in main run_all(args.n, args.batch_size) File "./cfg_sample.py", line 136, in run_all steps = utils.get_spliced_ddpm_cosine_schedule(t) File "C:\Users\m\Desktop\v-diffusion-pytorch\diffusion\utils.py", line 75, in get_spliced_ddpm_cosine_schedule ddpm_part = get_ddpm_schedule(big_t + ddpm_crossover - cosine_crossover) File "C:\Users\m\Desktop\v-diffusion-pytorch\diffusion\utils.py", line 65, in get_ddpm_schedule log_snr = -torch.special.expm1(1e-4 + 10 * ddpm_t**2).log() AttributeError: module 'torch' has no attribute 'special'

you need to create a token there https://pypi.org/manage/account/token/ and put it in there https://github.com/crowsonkb/v-diffusion-pytorch/settings/secrets/actions/new name it PYPI_PASSWORD

The release will be triggered when you name your commit Release x.y.z I advise to change the version in setup.cfg in that commit

Hello, thank you for your great work! I have a little puzzle in sample.py `# Get the model output (v, the predicted velocity) with torch.cuda.amp.autocast(): v = model(x, ts * steps[i], **extra_args).float()

    # Predict the noise and the denoised image
    pred = x * alphas[i] - v * sigmas[i]
    eps = x * sigmas[i] + v * alphas[i]`

what the meaning ? Where it comes?

Thanks for sharing such an amazing repo!

I am testing a prompt like openAI “an astronaut riding a horse in a photorealistic style” to compare. But somehow the iterations seems to be stuck on the same image.

This is my first test, so could very likely be that I am doing something wrong. Results and settings attached bellow…

Thanks for this great work. I'm recently interested in using diffusion model to generate images iteratively. I found your script cfg_sample.py was a nice implementation and I decided to learn from it. However, because I'm new in this field, I've encountered some problems quite hard to understand for me. It'd be great if some hints/suggestions are provided. Thank you!! My questions are listed below. They're about the script cfg_sample.py.

1. I noticed in the codes, we've used zero_embed as one of the features for conditioning. What is the purpose of using it? Is it designed to allow the case of no prompt for input?
2. I also notice that the weight of zero_embed is computed as 1 - sum(weights), I think the 1 is to make them sum to one, but actually the weight of zero_embed could be a negative number, should weights be normalized before all the intermediate noise maps are weighted?

Thanks very much!!

Hi!

I wanted to thank you for your work, especially since without you DiscoDiffusion wouldn't exist !

Still, I was wondering if you had the metrics (Precision, Recall, FID and Inception Score) on the 256x256 WikiArt model ?

Hey! love your work and been following your stuff for a while. I have finetuned a 64x64 unconditional model from openai/improved diffusion. checkpoint

I was curious if you could lend any insight on how to connect CLIP guidance to my model? I have tried repurposing your notebook (https://colab.research.google.com/drive/12a_Wrfi2_gwwAuN3VvMTwVMz9TfqctNj#scrollTo=1YwMUyt9LHG1) however past 100 steps, my models seems to unconverge.

I think perhaps because there is too much noise being added for the smaller image size? How might i fix this?