いつも通りdanbooruやreluとかで画像集めたらエロ画像のデータがやたら増えた普通に治療中の画像とかは割とセンシティブな個人情報だから集まり悪いのもうなづけるv002を作るにあたっていい画像のオススメがあれば教えて<h2 id="v001-tq5vho2p1">v001</h2>タグの手動整備なし10400step 100ep

点滴 intravenous drip-XLPony

# 点滴 v001### コマンドライン設定- <code>accelerate.EXE launch --dynamo_backend no --dynamo_mode default --mixed_precision bf16 --num_processes 2 --num_machines 1 --num_cpu_threads_per_process 2</code>- <code>sdxl_train_network.py --config_file config_lora-20240618-184456.toml</code>- <code>--lr_scheduler_type CosineAnnealingLR</code>- <code>--lr_scheduler_args T_max=1040 eta_min=1e-7</code>- <code>--network_args loraplus_lr_ratio=16</code>- <code>--alpha_mask</code>### toml設定ファイル- <code>bucket_no_upscale = true</code>- <code>bucket_reso_steps = 64</code>- <code>cache_latents = true</code>- <code>cache_latents_to_disk = true</code>- <code>caption_dropout_rate = 0.05</code>- <code>caption_extension = ".txt"</code>- <code>clip_skip = 1</code>- <code>dynamo_backend = "no"</code>- <code>enable_bucket = true</code>- <code>epoch = 100</code>- <code>gradient_accumulation_steps = 1</code>- <code>gradient_checkpointing = true</code>- <code>huber_c = 0.1</code>- <code>huber_schedule = "snr"</code>- <code>keep_tokens = 1</code>- <code>learning_rate = 0.0001</code>- <code>log_with = "wandb"</code>- <code>logging_dir = "log"</code>- <code>loss_type = "smooth_l1"</code>- <code>lr_scheduler = "cosine"</code>- <code>lr_scheduler_args = []</code>- <code>lr_scheduler_num_cycles = 1</code>- <code>lr_scheduler_power = 1</code>- <code>max_bucket_reso = 2048</code>- <code>max_data_loader_n_workers = 0</code>- <code>max_grad_norm = 0.1</code>- <code>max_timestep = 1000</code>- <code>max_token_length = 225</code>- <code>max_train_epochs = 100</code>- <code>max_train_steps = 10400</code>- <code>mem_eff_attn = true</code>- <code>min_bucket_reso = 256</code>- <code>min_snr_gamma = 5</code>- <code>mixed_precision = "bf16"</code>- <code>network_alpha = 32</code>- <code>network_args = ["rank_dropout=0.2", "module_dropout=0.2"]</code>- <code>network_dim = 32</code>- <code>network_dropout = 0.2</code>- <code>network_module = "networks.lora"</code>- <code>no_half_vae = true</code>- <code>noise_offset_type = "Original"</code>- <code>optimizer_args = ["weight_decay=0.01"]</code>- <code>optimizer_type = "AdamW8bit"</code>- <code>output_dir = "output"</code>- <code>output_name = "intravenousdrip-XLPony_001"</code>- <code>pretrained_model_name_or_path = "ponyDiffusionV6XL_v6StartWithThisOne.safetensors"</code>- <code>prior_loss_weight = 0.1</code>- <code>resolution = "1024,1024"</code>- <code>sample_prompts = "prompt.txt"</code>- <code>sample_sampler = "euler_a"</code>- <code>save_every_n_epochs = 10</code>- <code>save_model_as = "safetensors"</code>- <code>save_precision = "bf16"</code>- <code>save_state = true</code>- <code>seed = 1</code>- <code>shuffle_caption = true</code>- <code>text_encoder_lr = 1e-5</code>- <code>train_batch_size = 1</code>- <code>unet_lr = 0.0001</code>- <code>xformers = true</code>

Files with SHA256: 76e1b2b44743c15af0dc4a6e753a069dc656ae28026eb0d5ba4b431e464598ff