|
--- |
|
license: other |
|
license_name: fair-ai-public-license-1.0-sd |
|
license_link: https://freedevproject.org/faipl-1.0-sd/ |
|
datasets: |
|
- pls2000/aiart_channel_nai3_geachu |
|
base_model: |
|
- OnomaAIResearch/Illustrious-xl-early-release-v0 |
|
tags: |
|
- lora |
|
--- |
|
|
|
# Lora Training (`arcain_2411.safetensors`) |
|
Lora trained on Illustrious-xl v0.1, but this lora can applied with other ILXL-based models such as NoobAI-XL. |
|
|
|
- Tool: kohya-ss/sd-scripts |
|
- GPUs: 4x RTX3060 |
|
- Dataset: pls2000/aiart_channel_nai3_geachu + additional data until 24/11/14 - blue archive data |
|
- Time taken: 50.5 hours (walltime) |
|
|
|
|
|
#### lora_arcain.sh |
|
``` |
|
NCCL_P2P_DISABLE=1 NCCL_IB_DISABLE=1 accelerate launch --num_cpu_threads_per_process 4 sdxl_train_network.py \ |
|
--network_train_unet_only \ |
|
--network_module="networks.lora" --network_dim 128 --network_alpha 128 \ |
|
--pretrained_model_name_or_path="/ai/data/sd/models/Stable-diffusion/SDXL/Illustrious-XL-v0.1.safetensors" \ |
|
--dataset_config="arcain.lora.toml" \ |
|
--output_dir="results/lora" --output_name="arcain-`date +%y%m`" \ |
|
--save_model_as="safetensors" \ |
|
--train_batch_size 2 --gradient_accumulation_steps 64 \ |
|
--learning_rate=1e-5 --optimizer_type="Lion8bit" \ |
|
--lr_scheduler="constant_with_warmup" --lr_warmup_steps 100 --optimizer_args "weight_decay=0.01" "betas=0.9,0.95" --min_snr_gamma 5 \ |
|
--sdpa \ |
|
--no_half_vae \ |
|
--cache_latents --cache_latents_to_disk \ |
|
--gradient_checkpointing \ |
|
--full_bf16 --mixed_precision="bf16" --save_precision="fp16" \ |
|
--ddp_timeout=10000000 \ |
|
--max_train_epochs 8 --save_every_n_epochs 1 \ |
|
--log_with wandb --log_tracker_name kohya-ss --wandb_run_name "arcain_`date +%y%m%d-%H%M`" --logging_dir wandb \ |
|
``` |
|
|
|
#### arcain.lora.toml |
|
``` |
|
[general] |
|
shuffle_caption = true |
|
caption_tag_dropout_rate = 0.2 |
|
keep_tokens_separator = "|||" |
|
caption_extension = ".txt" |
|
|
|
[[datasets]] |
|
enable_bucket = true |
|
min_bucket_reso = 512 |
|
max_bucket_reso = 4096 |
|
resolution = 1024 |
|
[[datasets.subsets]] |
|
image_dir = "/mnt/wd8tb/train/to_train" |
|
num_repeats = 1 |
|
``` |
