Hello everyone. I am Dr. Furkan Gözükara. PhD Computer Engineer. SECourses is a dedicated YouTube channel for the following topics : Tech, AI, News, Science, Robotics, Singularity, ComfyUI, SwarmUI, ML, Artificial Intelligence, Humanoid Robots, Wan 2.2, FLUX, Krea, Qwen Image, VLMs, Stable Diffusion
I had to modify this code to generate TensorRT that will support highres fix for SDXL self.max_image_shape = 768 if version in ("1.4", "1.5") else 2048 It is by default 1024 So ...
@Dr. Furkan Gözükara Do you know if there is any known issue with Torch 2.1.0? I always see the dependencies pointing to 2.0.1. I did a test with 2.1.0 on my dreambooth 2.1.0 and noticed that it performed faster.
also interested in this topic since I tried an update from torch 2.0.1 to 2.1.1 with xformers 0.23 and couldn't get it to work. Always trying to figure out which dependencies are needed.
Well, I'm going to test Kohya with its default configuration, after I get good results I'll do the experiment and try to update the dependencies to the most recent ones.
Dr, you think we can train this new realistic SDXL model (https://civitai.com/models/152525) with your config files as base? the model contains VAE baked inside already
@Dr. Furkan Gözükara when you say that not using xformers with 24gb is better than using xformers with 16gb, does it mean that it only changes the quality of image rendering or affects the quality of training the face correctly?