diffusion-trainer/config_decoder.json at main · Disty0/diffusion-trainer · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
{
    "model_path": "/mnt/DataSSD/AI/models/test/sd3/32_sb-3.6B_raifu_plus",
    "model_type": "sd3",
    "sensitive_keys": [],
    "override_sensitive_keys": false,
    "optimizer": "torch.optim.AdamW",
    "optimizer_args": {"weight_decay": 0.01, "eps": 1e-8, "betas": [0.9, 0.999], "fused": true},
    "optimizer_args_sensitive": {"weight_decay": 0.0, "eps": 1e-8, "betas": [0.9, 0.999], "fused": true},
    "fused_optimizer": false,
    "optimizer_cpu_offload": false,
    "optimizer_offload_gradients": false,
    "lr_scheduler": "CosineAnnealingLR",
    "lr_scheduler_args": {"T_max": 102400, "eta_min": 4e-7},
    "epochs": 10,
    "learning_rate": 4e-6,
    "learning_rate_sensitive": 4e-8,
    "max_grad_norm": 0.0,
    "max_grad_clip": 0.0,
    "log_grad_stats": false,
    "loss_type": "mse",
    "loss_reduction": "mean",
    "batch_size": 1,
    "gradient_accumulation_steps": 16,
    "gradient_checkpointing": true,
    "weights_dtype": "float32",
    "mixed_precision": "fp16",
    "use_grad_scaler": false,
    "dynamo_backend": "no",
    "tunableop": null,
    "cudnn_enabled": true,
    "allow_tf32": true,
    "allow_reduced_precision": true,
    "flash_sdp": true,
    "mem_efficient_sdp": true,
    "dynamic_sdp": false,
    "math_sdp": true,
    "math_sdp_reduction": true,
    "ema_decay": 0.995,
    "ema_update_steps": 32,
    "use_ema": true,
    "use_foreach_ema": true,
    "update_ema_on_cpu": false,
    "offload_ema_to_cpu": false,
    "offload_ema_non_blocking": true,
    "offload_ema_pin_memory": false,
    "ema_weights_dtype": "float32",
    "gc_steps": 256,
    "checkpoint_save_steps": 256,
    "checkpoints_limit": 20,
    "load_queue_lenght": 64,
    "dataloader_pin_memory": false,
    "max_load_workers": 4,
    "reshuffle": true,
    "do_file_check": true,
    "log_with": "wandb",
    "project_name": "anime_decoder",
    "resume_from": "none",
    "dataset_index": "cache/decoder_dataset_index",
    "project_dir": "/mnt/DataSSD/AI/models/SoteDiffusion/StableDiffusion3/train_decoder",
    "image_ext": ".jxl",
    "dataset_paths": [
        {
            "path": "/mnt/DataSSD/AI/anime_image_dataset/vae/danbooru-best-png-latent",
            "bucket_list": "bucket_list.json",
            "images": "/mnt/DataSSD/AI/anime_image_dataset/vae/danbooru-best-png-cropped",
            "repeats": 1
        }
    ]
}