File size: 3,679 Bytes
ea718e2
9902b4c
ea718e2
9902b4c
ea718e2
 
9902b4c
ea718e2
9902b4c
 
6309d29
 
ea718e2
 
6309d29
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9902b4c
 
ea718e2
 
9902b4c
ea718e2
9902b4c
 
 
ea718e2
 
9902b4c
 
 
 
 
6309d29
ea718e2
9902b4c
 
 
ea718e2
9902b4c
 
 
6309d29
9902b4c
 
 
 
 
ea718e2
9902b4c
 
 
 
ea718e2
9902b4c
 
ea718e2
9902b4c
 
 
 
ea718e2
9902b4c
ea718e2
9902b4c
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
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
job: custom_job
config:
  name: flux_train_replicate
  process:
  - type: custom_sd_trainer
    training_folder: output
    device: cuda:0
    trigger_word: HST
    network:
      type: lora
      linear: 32
      linear_alpha: 32
      network_kwargs:
        only_if_contains:
        - transformer.transformer_blocks.7.norm1.linear
        - transformer.transformer_blocks.7.norm1_context.linear
        - transformer.transformer_blocks.7.attn.to_q
        - transformer.transformer_blocks.7.attn.to_k
        - transformer.transformer_blocks.7.attn.to_v
        - transformer.transformer_blocks.7.attn.add_k_proj
        - transformer.transformer_blocks.7.attn.add_v_proj
        - transformer.transformer_blocks.7.attn.add_q_proj
        - transformer.transformer_blocks.7.attn.to_out.0
        - transformer.transformer_blocks.7.attn.to_add_out
        - transformer.transformer_blocks.7.ff.net.0.proj
        - transformer.transformer_blocks.7.ff.net.2
        - transformer.transformer_blocks.7.ff_context.net.0.proj
        - transformer.transformer_blocks.7.ff_context.net.2
        - transformer.transformer_blocks.13.norm1.linear
        - transformer.transformer_blocks.13.norm1_context.linear
        - transformer.transformer_blocks.13.attn.to_q
        - transformer.transformer_blocks.13.attn.to_k
        - transformer.transformer_blocks.13.attn.to_v
        - transformer.transformer_blocks.13.attn.add_k_proj
        - transformer.transformer_blocks.13.attn.add_v_proj
        - transformer.transformer_blocks.13.attn.add_q_proj
        - transformer.transformer_blocks.13.attn.to_out.0
        - transformer.transformer_blocks.13.attn.to_add_out
        - transformer.transformer_blocks.13.ff.net.0.proj
        - transformer.transformer_blocks.13.ff.net.2
        - transformer.transformer_blocks.13.ff_context.net.0.proj
        - transformer.transformer_blocks.13.ff_context.net.2
        - transformer.single_transformer_blocks.7.norm.linear
        - transformer.single_transformer_blocks.7.proj_mlp
        - transformer.single_transformer_blocks.7.proj_out
        - transformer.single_transformer_blocks.7.attn.to_q
        - transformer.single_transformer_blocks.7.attn.to_k
        - transformer.single_transformer_blocks.7.attn.to_v
        - transformer.single_transformer_blocks.13.norm.linear
        - transformer.single_transformer_blocks.13.proj_mlp
        - transformer.single_transformer_blocks.13.proj_out
        - transformer.single_transformer_blocks.13.attn.to_q
        - transformer.single_transformer_blocks.13.attn.to_k
        - transformer.single_transformer_blocks.13.attn.to_v
    save:
      dtype: float16
      save_every: 501
      max_step_saves_to_keep: 1
    datasets:
    - folder_path: input_images
      caption_ext: txt
      caption_dropout_rate: 0.05
      shuffle_tokens: false
      cache_latents_to_disk: false
      cache_latents: true
      resolution:
      - 512
      - 768
      - 1024
    train:
      batch_size: 4
      steps: 500
      gradient_accumulation_steps: 1
      train_unet: true
      train_text_encoder: false
      content_or_style: balanced
      gradient_checkpointing: true
      noise_scheduler: flowmatch
      optimizer: adamw8bit
      lr: 0.001
      ema_config:
        use_ema: true
        ema_decay: 0.99
      dtype: bf16
    model:
      name_or_path: FLUX.1-dev
      is_flux: true
      quantize: true
    sample:
      sampler: flowmatch
      sample_every: 501
      width: 1024
      height: 1024
      prompts: []
      neg: ''
      seed: 42
      walk_seed: true
      guidance_scale: 3.5
      sample_steps: 28
meta:
  name: flux_train_replicate
  version: '1.0'