克里斯提娜 Christina

Lora 2 年前专属

Trained on this collab https://colab.research.google.com/drive/1bFX0pZczeApeFadrz1AdOb5TDdet2U0Z?usp=sharing

With these parameters:

!accelerate launch \ –num_cpu_threads_per_process 12 train_network.py \ –network_module=networks.lora \ –pretrained_model_name_or_path=”/content/model/nai.ckpt” \ –train_data_dir=”/content/drive/MyDrive/sd/dataset/sincos” \ –reg_data_dir=”/content/drive/MyDrive/sd/dataset/sincos_reg” \ –output_dir=”/content/drive/MyDrive/sd/stable-diffusion-webui/models/Lora/” \ –caption_extension=”.txt” \ –prior_loss_weight=1 \ –resolution=768 \ –enable_bucket \ –min_bucket_reso=320 \ –max_bucket_reso=960 \ –train_batch_size=3 \ –learning_rate=1e-5 \ –unet_lr=1.5e-4 \ –text_encoder_lr=1.5e-5 \ –max_train_epochs=10 \ –mixed_precision=”fp16″ \ –save_precision=”fp16″ \ –use_8bit_adam \ –xformers \ –save_every_n_epochs=1\ –save_model_as=safetensors \ –clip_skip=2 \ –seed=31337 \ –network_dim=128 \ –network_alpha=128 \ –max_token_length=150 \ –cache_latents \ –lr_scheduler=”constant_with_warmup” \ –lr_warmup=5 \ –lr_scheduler_num_cycle=1 \ –output_name=”christina” \ –shuffle_caption \ –keep_tokens=1 \ –max_data_loader_n_workers=1 \ –persistent_data_loader_workers \ –bucket_no_upscale \ –bucket_reso_steps=64 \ # –lr_warmup_steps= 40 \ # –noise_offset=0.1 \

Here is the tag_frequency: https://files.catbox.moe/4ix2me.txt
For activation tokens check the 5th image.

Hide

声明：本站仅供资源信息交流学习，所有资源均为网络收集整理不保证内容完整性，仅供学习和研究使用。如侵犯到您的权益，请联系本站下架。