kohya_ss_anima/library/class_basic_training.py at 85f8ece7c9a65df368bd46b24b7782d9495e8876 · Source2Spy/kohya_ss_anima · 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
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
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
import gradio as gr
import os

class BasicTraining:
    def __init__(
        self,
        learning_rate_value='1e-6',
        lr_scheduler_value='constant',
        lr_warmup_value='0',
        finetuning: bool = False,
    ):
        self.learning_rate_value = learning_rate_value
        self.lr_scheduler_value = lr_scheduler_value
        self.lr_warmup_value = lr_warmup_value
        self.finetuning = finetuning

        with gr.Row():
            self.train_batch_size = gr.Slider(
                minimum=1,
                maximum=64,
                label='Train batch size',
                value=1,
                step=1,
            )
            self.epoch = gr.Number(label='Epoch', value=1, precision=0)
            self.max_train_epochs = gr.Textbox(
                label='Max train epoch',
                placeholder='(Optional) Enforce number of epoch',
            )
            self.save_every_n_epochs = gr.Number(
                label='Save every N epochs', value=1, precision=0
            )
            self.caption_extension = gr.Textbox(
                label='Caption Extension',
                placeholder='(Optional) Extension for caption files. default: .caption',
            )
        with gr.Row():
            self.mixed_precision = gr.Dropdown(
                label='Mixed precision',
                choices=[
                    'no',
                    'fp16',
                    'bf16',
                ],
                value='fp16',
            )
            self.save_precision = gr.Dropdown(
                label='Save precision',
                choices=[
                    'float',
                    'fp16',
                    'bf16',
                ],
                value='fp16',
            )
            self.num_cpu_threads_per_process = gr.Slider(
                minimum=1,
                maximum=os.cpu_count(),
                step=1,
                label='Number of CPU threads per core',
                value=2,
            )
            self.seed = gr.Textbox(
                label='Seed', placeholder='(Optional) eg:1234'
            )
            self.cache_latents = gr.Checkbox(label='Cache latents', value=True)
            self.cache_latents_to_disk = gr.Checkbox(
                label='Cache latents to disk', value=False
            )
        with gr.Row():
            self.learning_rate = gr.Number(
                label='Learning rate', value=learning_rate_value
            )
            self.lr_scheduler = gr.Dropdown(
                label='LR Scheduler',
                choices=[
                    'adafactor',
                    'constant',
                    'constant_with_warmup',
                    'cosine',
                    'cosine_with_restarts',
                    'linear',
                    'polynomial',
                ],
                value=lr_scheduler_value,
            )
            self.lr_warmup = gr.Slider(
                label='LR warmup (% of steps)',
                value=lr_warmup_value,
                minimum=0,
                maximum=100,
                step=1,
            )
            self.optimizer = gr.Dropdown(
                label='Optimizer',
                choices=[
                    'AdamW',
                    'AdamW8bit',
                    'Adafactor',
                    'DAdaptation',
                    'DAdaptAdaGrad',
                    'DAdaptAdam',
                    'DAdaptAdan',
                    'DAdaptAdanIP',
                    'DAdaptAdamPreprint',
                    'DAdaptLion',
                    'DAdaptSGD',
                    'Lion',
                    'Lion8bit',
                    "PagedAdamW8bit",
                    "PagedLion8bit",
                    'Prodigy',
                    'SGDNesterov',
                    'SGDNesterov8bit',
                ],
                value='AdamW8bit',
                interactive=True,
            )
        with gr.Row(visible=not finetuning):
            self.lr_scheduler_num_cycles = gr.Textbox(
                label='LR number of cycles',
                placeholder='(Optional) For Cosine with restart and polynomial only',
            )

            self.lr_scheduler_power = gr.Textbox(
                label='LR power',
                placeholder='(Optional) For Cosine with restart and polynomial only',
            )
        with gr.Row():
            self.optimizer_args = gr.Textbox(
                label='Optimizer extra arguments',
                placeholder='(Optional) eg: relative_step=True scale_parameter=True warmup_init=True',
            )
        with gr.Row(visible=not finetuning):
            self.max_resolution = gr.Textbox(
                label='Max resolution',
                value='512,512',
                placeholder='512,512',
            )
            self.stop_text_encoder_training = gr.Slider(
                minimum=-1,
                maximum=100,
                value=0,
                step=1,
                label='Stop text encoder training',
            )
        with gr.Row(visible=not finetuning):
            self.enable_bucket = gr.Checkbox(label='Enable buckets', value=True)
            self.min_bucket_reso = gr.Slider(label='Minimum bucket resolution', value=256, minimum=64, maximum=4096, step=64, info='Minimum size in pixel a bucket can be (>= 64)')
            self.max_bucket_reso = gr.Slider(label='Maximum bucket resolution', value=2048, minimum=64, maximum=4096, step=64, info='Maximum size in pixel a bucket can be (>= 64)')