File size: 6,987 Bytes
0a82683
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
import torch


def load_text_encoders(args, class_one, class_two):
    text_encoder_one = class_one.from_pretrained(
        args.pretrained_model_name_or_path, subfolder="text_encoder", revision=args.revision, variant=args.variant
    )
    text_encoder_two = class_two.from_pretrained(
        args.pretrained_model_name_or_path, subfolder="text_encoder_2", revision=args.revision, variant=args.variant
    )
    return text_encoder_one, text_encoder_two


def tokenize_prompt(tokenizer, prompt, max_sequence_length):
    text_inputs = tokenizer(
        prompt,
        padding="max_length",
        max_length=max_sequence_length,
        truncation=True,
        return_length=False,
        return_overflowing_tokens=False,
        return_tensors="pt",
    )
    text_input_ids = text_inputs.input_ids
    return text_input_ids


def tokenize_prompt_clip(tokenizer, prompt):
    text_inputs = tokenizer(
        prompt,
        padding="max_length",
        max_length=77,
        truncation=True,
        return_length=False,
        return_overflowing_tokens=False,
        return_tensors="pt",
    )
    text_input_ids = text_inputs.input_ids
    return text_input_ids


def tokenize_prompt_t5(tokenizer, prompt):
    text_inputs = tokenizer(
        prompt,
        padding="max_length",
        max_length=512,
        truncation=True,
        return_length=False,
        return_overflowing_tokens=False,
        return_tensors="pt",
    )
    text_input_ids = text_inputs.input_ids
    return text_input_ids


def _encode_prompt_with_t5(
        text_encoder,
        tokenizer,
        max_sequence_length=512,
        prompt=None,
        num_images_per_prompt=1,
        device=None,
        text_input_ids=None,
):
    prompt = [prompt] if isinstance(prompt, str) else prompt
    batch_size = len(prompt)

    if tokenizer is not None:
        text_inputs = tokenizer(
            prompt,
            padding="max_length",
            max_length=max_sequence_length,
            truncation=True,
            return_length=False,
            return_overflowing_tokens=False,
            return_tensors="pt",
        )
        text_input_ids = text_inputs.input_ids
    else:
        if text_input_ids is None:
            raise ValueError("text_input_ids must be provided when the tokenizer is not specified")

    prompt_embeds = text_encoder(text_input_ids.to(device))[0]

    dtype = text_encoder.dtype
    prompt_embeds = prompt_embeds.to(dtype=dtype, device=device)

    _, seq_len, _ = prompt_embeds.shape

    # duplicate text embeddings and attention mask for each generation per prompt, using mps friendly method
    prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
    prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)

    return prompt_embeds


def _encode_prompt_with_clip(
        text_encoder,
        tokenizer,
        prompt: str,
        device=None,
        text_input_ids=None,
        num_images_per_prompt: int = 1,
):
    prompt = [prompt] if isinstance(prompt, str) else prompt
    batch_size = len(prompt)

    if tokenizer is not None:
        text_inputs = tokenizer(
            prompt,
            padding="max_length",
            max_length=77,
            truncation=True,
            return_overflowing_tokens=False,
            return_length=False,
            return_tensors="pt",
        )

        text_input_ids = text_inputs.input_ids
    else:
        if text_input_ids is None:
            raise ValueError("text_input_ids must be provided when the tokenizer is not specified")

    prompt_embeds = text_encoder(text_input_ids.to(device), output_hidden_states=False)

    # Use pooled output of CLIPTextModel
    prompt_embeds = prompt_embeds.pooler_output
    prompt_embeds = prompt_embeds.to(dtype=text_encoder.dtype, device=device)

    # duplicate text embeddings for each generation per prompt, using mps friendly method
    prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
    prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, -1)

    return prompt_embeds


def encode_prompt(
        text_encoders,
        tokenizers,
        prompt: str,
        max_sequence_length,
        device=None,
        num_images_per_prompt: int = 1,
        text_input_ids_list=None,
):
    prompt = [prompt] if isinstance(prompt, str) else prompt
    dtype = text_encoders[0].dtype

    pooled_prompt_embeds = _encode_prompt_with_clip(
        text_encoder=text_encoders[0],
        tokenizer=tokenizers[0],
        prompt=prompt,
        device=device if device is not None else text_encoders[0].device,
        num_images_per_prompt=num_images_per_prompt,
        text_input_ids=text_input_ids_list[0] if text_input_ids_list else None,
    )

    prompt_embeds = _encode_prompt_with_t5(
        text_encoder=text_encoders[1],
        tokenizer=tokenizers[1],
        max_sequence_length=max_sequence_length,
        prompt=prompt,
        num_images_per_prompt=num_images_per_prompt,
        device=device if device is not None else text_encoders[1].device,
        text_input_ids=text_input_ids_list[1] if text_input_ids_list else None,
    )

    text_ids = torch.zeros(prompt_embeds.shape[1], 3).to(device=device, dtype=dtype)

    return prompt_embeds, pooled_prompt_embeds, text_ids


def encode_token_ids(text_encoders, tokens, accelerator, num_images_per_prompt=1, device=None):
    text_encoder_clip = text_encoders[0]
    text_encoder_t5 = text_encoders[1]
    tokens_clip, tokens_t5 = tokens[0], tokens[1]
    batch_size = tokens_clip.shape[0]

    if device == "cpu":
        device = "cpu"
    else:
        device = accelerator.device

    # clip
    prompt_embeds = text_encoder_clip(tokens_clip.to(device), output_hidden_states=False)
    # Use pooled output of CLIPTextModelpreprocess_train
    prompt_embeds = prompt_embeds.pooler_output
    prompt_embeds = prompt_embeds.to(dtype=text_encoder_clip.dtype, device=accelerator.device)
    # duplicate text embeddings for each generation per prompt, using mps friendly method
    prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
    pooled_prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, -1)
    pooled_prompt_embeds = pooled_prompt_embeds.to(dtype=text_encoder_clip.dtype, device=accelerator.device)

    # t5
    prompt_embeds = text_encoder_t5(tokens_t5.to(device))[0]
    dtype = text_encoder_t5.dtype
    prompt_embeds = prompt_embeds.to(dtype=dtype, device=accelerator.device)
    _, seq_len, _ = prompt_embeds.shape
    # duplicate text embeddings and attention mask for each generation per prompt, using mps friendly method
    prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
    prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)

    text_ids = torch.zeros(prompt_embeds.shape[1], 3).to(device=accelerator.device, dtype=dtype)

    return prompt_embeds, pooled_prompt_embeds, text_ids