André Longon

I initially also had trouble getting pretrained weights working properly with this repository but I resolved the problem. Not sure if you have the same problem, but I'll relay in...

@klei22 Apologies, I did rename the pth files so that may have added confusion. For prior, I am using latest.pth and prior_config.json in huggingface's prior folder. For decoder, I am...

@tikitong Sure thing. A reminder that this fix requires downgrading dalle2-pytorch to version 1.1.0. Could you be more specific about what doesn't work on your end? Do you get an...

@ZhangxinruBIT You didn't mention changing the keys in the decoder. This was something I mentioned and included in the code. ``` for k in decoder.clip.state_dict().keys(): decoder_model_state["clip." + k] = decoder.clip.state_dict()[k]...

@tikitong I think you've got it working. The fact that you have a clean image that's at least car related makes me think it's working properly. The imperfect results are...

@kdavidlp123 You're not using the exact same code as you're trying to set your prior and decoder to your cpu. I've only ran everything on GPU so not sure what...

@kdavidlp123 Open the decoder_config json and change "image_sizes". I was having trouble setting it to a higher value. The images would look very oddly distorted. I don't know much about...

> @cest-andre can you please attach the updated code here for others to use as an initial reference, that will be a great help. The bottom of my first comment...

#282 See my response here. This may help you too.

I've found a solution for purposes, and hopefully it works for anyone having this problem. Looks like the device is hardcoded to be "cuda:0" in various parts of the code,...