trufty

Old workaround, but should still be valid https://github.com/borisdayma/dalle-mini/issues/257#issuecomment-1157198565

Yea you can get both VQGan, and Dalle-mega models below. (don't forget the config json files) Current mega version is 14 at this moment. > https://huggingface.co/dalle-mini/vqgan_imagenet_f16_16384 > https://wandb.ai/dalle-mini/dalle-mini/artifacts/DalleBart_model/mega-1-fp16/46cca62fd0ec6fcab873/files Download them...

Not sure about that one.

I just altered the `seed` code so I can either enter a custom seed, or use a random one. Obviously if you use a static seed like below, all of...

I just reverted my changes to test the vanilla implementation, and it seems to be working as expected. Though I'm not sure the seed is ever logged anywhere for the...

A little late, but wanted to drop in and say I'm running fine at 512x512 with a GTX 1080 8GB vram with version 1.5.1 of this repository. Currently on 39k...

Edit: just noticed you were the reporter! Thanks for that. Looks like pytorch-fid just fixed this in their [repo](https://github.com/mseitzer/pytorch-fid/commit/732341f2129bc8b41cbbfaa4d738788645cb1df7) Not sure if its available in pip yet.

I'm getting the same thing, but that looks like a deep rabbit hole. I just ended up downloading this repo and training via `python ./stylegan2_pytorch/cli.py ` You'll have to alter...

I was also getting `NaN` loss after a few thousand steps on versions 0.25.3 AND 0.26.1. This was happening with both `GaussianDiffusion` and `ElucidatedDiffusion` For me the solution was setting...

12f95b33d8a2f44ffb48ec1079083ec634c4728f seems to have fixed this issue for me. I can set `amp = True` again without NaN.