ShivamShrirao
train_dreambooth_lora.py on Colab
Describe the bug
I'm trying to run this script, starting with swapping the default train_dreambooth.py with the lora one in install requirements and run cells. Then I modify the settings to remove all the arguments lora doesn't use and adding the ones it needs, such as instance and class dirs and prompts, since it doesn't use concepts_list. I also change the learning rate to 1e-4. I find that I am able to complete the training, it saves the file and checkpoints but I am not seeing results in inference.
I don't know if it's working. Either the training is not working out or I am running inference the wrong way... To the inference cell I added: pipe.unet.load_attn_procs(model_path) pipe.to("cuda")
With the right path to embeds, but am just not seeing results.
Anyone have any suggestions?
Reproduction
null
Logs
No response
System Info
https://colab.research.google.com/github/ShivamShrirao/diffusers/blob/main/examples/dreambooth/DreamBooth_Stable_Diffusion.ipynb
the results in Lora training saves the images to the log folder and can only be seen in WandB. I hate that and i hate tensorboard and all of that. thats why i never train lora lol. you could try finding where it saves the weights and add some lines to make it generate 4 images and display them
