Brett Kuprel
Brett Kuprel
Each mini image is 256x256 and can't be changed. You can make the grid size bigger though or use an upscaling model like https://replicate.com/jingyunliang/swinir
Thanks for doing this. I actually just added a --grid-size flag, does that work for you?
I merged your pull request that adds the line `torch.set_num_threads(os.cpu_count())` but it still doesn't seem to be using all the threads on my M1 macbook. I agree there should be...
Awesome!
I was also looking into getting this model on the phone. Apple says that for transformers in pytorch, the dimensions aren't in optimal order for the neural engine: https://machinelearning.apple.com/research/neural-engine-transformers They...
Wow cool! I'm getting a 404 error when I click your link to the app
I'm not sure what you mean. Are you saying parallel forward over the 256 image tokens? That wouldn't work because each token depends on the previous token. And if you...
Oh I see, it would be for if you wanted to do a forward pass over all tokens at once, instead of sampling one after the other.
What are the advantages over using requests?
Wow this is great! I just added your video to the readme. You're right the clamping is unnecessary. It originally served to avoid a cryptic cuda runtime error. Later I...