sd-scripts
sd-scripts copied to clipboard
kohya-ss
`RuntimeError: Input type (float) and bias type (c10::BFloat16) should be the same` occupied. when I set mixed_precision="no", it works instead
Majorly 2 places has been modified 1, in accelerate multi-gpu mode, for this sd-scripts training, has to make sure all child-process are using the same random_seed, because in the training...
Here are examples of samples that was generated when i trained a short test dreambooth for just 400 steps(sample every 100 steps)  But when i ran the the minimal...
im running into this error when starting the training. I have renamed all my files in case it was something with the filenames causing issues and it did not resolve...
1.i have install deepspeed with pip 2.this is my train script: accelerate launch --num_cpu_threads_per_process=2 train_network.py \ --pretrained_model_name_or_path="xxx" \ --train_data_dir="xxx" \ --caption_extension='.txt' \ --resolution="840,600" \ --tokenizer_cache_dir="xxx" \ --logging_dir="xxx" \ --output_dir="xxx" \...
Hello, I seem to be getting the below error, just to point out I've tried using BF16 but it just hangs on caching latents, I've also tried changing the optimiser...
This PR adds support for training with [token downsampling](https://github.com/ethansmith2000/ImprovedTokenMerge) and replaces my token merge PR (#1146). Token downsampling is a lossy optimization that significantly speeds up inference and training. It...
I'm training a SDXL Lora and I don't understand why some of my images end up in the 960x960 bucket. Shouldn't the square and square like images go to the...
Hey @kohya-ss and @BootsofLagrangian Thank you for your hard work on Deepspeed integration! As I see this feature in dev branch. I tried to use deepspeed in my pipelines but...
Based on PR #914 "Add validation loss" by @rockerBOO works, this proposal aims to integrate the feature into the latest version with minimal modifications to the existing functionality. This update...
