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
@Dr. Furkan Gözükara thanks for the response, but something is not right I did this training many times and I never seen a base model to get so cooked so fast, 30 repetitions 320 imágenes and the model brakes , may be is my dataset and nobregularization but probabably for my case I need to reduce de learning rate or something , this images were used like a sanity check, thanks
I train locally on a 3090, I use your last 24gb slow preset, I was changing some parameters like crossatention for a speed up with no gradient checkpointing, I dont remember changing anything else
other questions did you tried this new learning rates on the base model for a run at least, may be works perfect on realvis4 but is too high for the base model
thanks for your great work, may be, be good if you make a run with your dataset, same settings and samples on the base model to be sure that is consistent on all scenarios.
Yes, the comfy ui tutorial you linked for me earlier ha. I think I'll wait for your video tutorial on how to upload lora safetensor from huggingface to runpod
hello Can you make one click for CodeFormer I am a photographer and I just want a sample gradio like the one huggingface I will be a Gold Supporter if you make it