Spaces:
Runtime error
Runtime error
File size: 16,394 Bytes
ad93086 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 |
# import math
# from collections import namedtuple
#
# import torch
#
# from modules import prompt_parser, devices, sd_hijack, sd_emphasis
# from modules.shared import opts
#
#
# class PromptChunk:
# """
# This object contains token ids, weight (multipliers:1.4) and textual inversion embedding info for a chunk of prompt.
# If a prompt is short, it is represented by one PromptChunk, otherwise, multiple are necessary.
# Each PromptChunk contains an exact amount of tokens - 77, which includes one for start and end token,
# so just 75 tokens from prompt.
# """
#
# def __init__(self):
# self.tokens = []
# self.multipliers = []
# self.fixes = []
#
#
# PromptChunkFix = namedtuple('PromptChunkFix', ['offset', 'embedding'])
# """An object of this type is a marker showing that textual inversion embedding's vectors have to placed at offset in the prompt
# chunk. Those objects are found in PromptChunk.fixes and, are placed into FrozenCLIPEmbedderWithCustomWordsBase.hijack.fixes, and finally
# are applied by sd_hijack.EmbeddingsWithFixes's forward function."""
#
#
# class TextConditionalModel(torch.nn.Module):
# def __init__(self):
# super().__init__()
#
# self.hijack = sd_hijack.model_hijack
# self.chunk_length = 75
#
# self.is_trainable = False
# self.input_key = 'txt'
# self.return_pooled = False
#
# self.comma_token = None
# self.id_start = None
# self.id_end = None
# self.id_pad = None
#
# def empty_chunk(self):
# """creates an empty PromptChunk and returns it"""
#
# chunk = PromptChunk()
# chunk.tokens = [self.id_start] + [self.id_end] * (self.chunk_length + 1)
# chunk.multipliers = [1.0] * (self.chunk_length + 2)
# return chunk
#
# def get_target_prompt_token_count(self, token_count):
# """returns the maximum number of tokens a prompt of a known length can have before it requires one more PromptChunk to be represented"""
#
# return math.ceil(max(token_count, 1) / self.chunk_length) * self.chunk_length
#
# def tokenize(self, texts):
# """Converts a batch of texts into a batch of token ids"""
#
# raise NotImplementedError
#
# def encode_with_transformers(self, tokens):
# """
# converts a batch of token ids (in python lists) into a single tensor with numeric representation of those tokens;
# All python lists with tokens are assumed to have same length, usually 77.
# if input is a list with B elements and each element has T tokens, expected output shape is (B, T, C), where C depends on
# model - can be 768 and 1024.
# Among other things, this call will read self.hijack.fixes, apply it to its inputs, and clear it (setting it to None).
# """
#
# raise NotImplementedError
#
# def encode_embedding_init_text(self, init_text, nvpt):
# """Converts text into a tensor with this text's tokens' embeddings. Note that those are embeddings before they are passed through
# transformers. nvpt is used as a maximum length in tokens. If text produces less teokens than nvpt, only this many is returned."""
#
# raise NotImplementedError
#
# def tokenize_line(self, line):
# """
# this transforms a single prompt into a list of PromptChunk objects - as many as needed to
# represent the prompt.
# Returns the list and the total number of tokens in the prompt.
# """
#
# if opts.emphasis != "None":
# parsed = prompt_parser.parse_prompt_attention(line)
# else:
# parsed = [[line, 1.0]]
#
# tokenized = self.tokenize([text for text, _ in parsed])
#
# chunks = []
# chunk = PromptChunk()
# token_count = 0
# last_comma = -1
#
# def next_chunk(is_last=False):
# """puts current chunk into the list of results and produces the next one - empty;
# if is_last is true, tokens <end-of-text> tokens at the end won't add to token_count"""
# nonlocal token_count
# nonlocal last_comma
# nonlocal chunk
#
# if is_last:
# token_count += len(chunk.tokens)
# else:
# token_count += self.chunk_length
#
# to_add = self.chunk_length - len(chunk.tokens)
# if to_add > 0:
# chunk.tokens += [self.id_end] * to_add
# chunk.multipliers += [1.0] * to_add
#
# chunk.tokens = [self.id_start] + chunk.tokens + [self.id_end]
# chunk.multipliers = [1.0] + chunk.multipliers + [1.0]
#
# last_comma = -1
# chunks.append(chunk)
# chunk = PromptChunk()
#
# for tokens, (text, weight) in zip(tokenized, parsed):
# if text == 'BREAK' and weight == -1:
# next_chunk()
# continue
#
# position = 0
# while position < len(tokens):
# token = tokens[position]
#
# if token == self.comma_token:
# last_comma = len(chunk.tokens)
#
# # this is when we are at the end of allotted 75 tokens for the current chunk, and the current token is not a comma. opts.comma_padding_backtrack
# # is a setting that specifies that if there is a comma nearby, the text after the comma should be moved out of this chunk and into the next.
# elif opts.comma_padding_backtrack != 0 and len(chunk.tokens) == self.chunk_length and last_comma != -1 and len(chunk.tokens) - last_comma <= opts.comma_padding_backtrack:
# break_location = last_comma + 1
#
# reloc_tokens = chunk.tokens[break_location:]
# reloc_mults = chunk.multipliers[break_location:]
#
# chunk.tokens = chunk.tokens[:break_location]
# chunk.multipliers = chunk.multipliers[:break_location]
#
# next_chunk()
# chunk.tokens = reloc_tokens
# chunk.multipliers = reloc_mults
#
# if len(chunk.tokens) == self.chunk_length:
# next_chunk()
#
# embedding, embedding_length_in_tokens = self.hijack.embedding_db.find_embedding_at_position(tokens, position)
# if embedding is None:
# chunk.tokens.append(token)
# chunk.multipliers.append(weight)
# position += 1
# continue
#
# emb_len = int(embedding.vectors)
# if len(chunk.tokens) + emb_len > self.chunk_length:
# next_chunk()
#
# chunk.fixes.append(PromptChunkFix(len(chunk.tokens), embedding))
#
# chunk.tokens += [0] * emb_len
# chunk.multipliers += [weight] * emb_len
# position += embedding_length_in_tokens
#
# if chunk.tokens or not chunks:
# next_chunk(is_last=True)
#
# return chunks, token_count
#
# def process_texts(self, texts):
# """
# Accepts a list of texts and calls tokenize_line() on each, with cache. Returns the list of results and maximum
# length, in tokens, of all texts.
# """
#
# token_count = 0
#
# cache = {}
# batch_chunks = []
# for line in texts:
# if line in cache:
# chunks = cache[line]
# else:
# chunks, current_token_count = self.tokenize_line(line)
# token_count = max(current_token_count, token_count)
#
# cache[line] = chunks
#
# batch_chunks.append(chunks)
#
# return batch_chunks, token_count
#
# def forward(self, texts):
# """
# Accepts an array of texts; Passes texts through transformers network to create a tensor with numerical representation of those texts.
# Returns a tensor with shape of (B, T, C), where B is length of the array; T is length, in tokens, of texts (including padding) - T will
# be a multiple of 77; and C is dimensionality of each token - for SD1 it's 768, for SD2 it's 1024, and for SDXL it's 1280.
# An example shape returned by this function can be: (2, 77, 768).
# For SDXL, instead of returning one tensor avobe, it returns a tuple with two: the other one with shape (B, 1280) with pooled values.
# Webui usually sends just one text at a time through this function - the only time when texts is an array with more than one element
# is when you do prompt editing: "a picture of a [cat:dog:0.4] eating ice cream"
# """
#
# batch_chunks, token_count = self.process_texts(texts)
#
# used_embeddings = {}
# chunk_count = max([len(x) for x in batch_chunks])
#
# zs = []
# for i in range(chunk_count):
# batch_chunk = [chunks[i] if i < len(chunks) else self.empty_chunk() for chunks in batch_chunks]
#
# tokens = [x.tokens for x in batch_chunk]
# multipliers = [x.multipliers for x in batch_chunk]
# self.hijack.fixes = [x.fixes for x in batch_chunk]
#
# for fixes in self.hijack.fixes:
# for _position, embedding in fixes:
# used_embeddings[embedding.name] = embedding
# devices.torch_npu_set_device()
# z = self.process_tokens(tokens, multipliers)
# zs.append(z)
#
# if opts.textual_inversion_add_hashes_to_infotext and used_embeddings:
# hashes = []
# for name, embedding in used_embeddings.items():
# shorthash = embedding.shorthash
# if not shorthash:
# continue
#
# name = name.replace(":", "").replace(",", "")
# hashes.append(f"{name}: {shorthash}")
#
# if hashes:
# if self.hijack.extra_generation_params.get("TI hashes"):
# hashes.append(self.hijack.extra_generation_params.get("TI hashes"))
# self.hijack.extra_generation_params["TI hashes"] = ", ".join(hashes)
#
# if any(x for x in texts if "(" in x or "[" in x) and opts.emphasis != "Original":
# self.hijack.extra_generation_params["Emphasis"] = opts.emphasis
#
# if self.return_pooled:
# return torch.hstack(zs), zs[0].pooled
# else:
# return torch.hstack(zs)
#
# def process_tokens(self, remade_batch_tokens, batch_multipliers):
# """
# sends one single prompt chunk to be encoded by transformers neural network.
# remade_batch_tokens is a batch of tokens - a list, where every element is a list of tokens; usually
# there are exactly 77 tokens in the list. batch_multipliers is the same but for multipliers instead of tokens.
# Multipliers are used to give more or less weight to the outputs of transformers network. Each multiplier
# corresponds to one token.
# """
# tokens = torch.asarray(remade_batch_tokens).to(devices.device)
#
# # this is for SD2: SD1 uses the same token for padding and end of text, while SD2 uses different ones.
# if self.id_end != self.id_pad:
# for batch_pos in range(len(remade_batch_tokens)):
# index = remade_batch_tokens[batch_pos].index(self.id_end)
# tokens[batch_pos, index+1:tokens.shape[1]] = self.id_pad
#
# z = self.encode_with_transformers(tokens)
#
# pooled = getattr(z, 'pooled', None)
#
# emphasis = sd_emphasis.get_current_option(opts.emphasis)()
# emphasis.tokens = remade_batch_tokens
# emphasis.multipliers = torch.asarray(batch_multipliers).to(devices.device)
# emphasis.z = z
#
# emphasis.after_transformers()
#
# z = emphasis.z
#
# if pooled is not None:
# z.pooled = pooled
#
# return z
#
#
# class FrozenCLIPEmbedderWithCustomWordsBase(TextConditionalModel):
# """A pytorch module that is a wrapper for FrozenCLIPEmbedder module. it enhances FrozenCLIPEmbedder, making it possible to
# have unlimited prompt length and assign weights to tokens in prompt.
# """
#
# def __init__(self, wrapped, hijack):
# super().__init__()
#
# self.hijack = hijack
#
# self.wrapped = wrapped
# """Original FrozenCLIPEmbedder module; can also be FrozenOpenCLIPEmbedder or xlmr.BertSeriesModelWithTransformation,
# depending on model."""
#
# self.is_trainable = getattr(wrapped, 'is_trainable', False)
# self.input_key = getattr(wrapped, 'input_key', 'txt')
# self.return_pooled = getattr(self.wrapped, 'return_pooled', False)
#
# self.legacy_ucg_val = None # for sgm codebase
#
# def forward(self, texts):
# if opts.use_old_emphasis_implementation:
# import modules.sd_hijack_clip_old
# return modules.sd_hijack_clip_old.forward_old(self, texts)
#
# return super().forward(texts)
#
#
# class FrozenCLIPEmbedderWithCustomWords(FrozenCLIPEmbedderWithCustomWordsBase):
# def __init__(self, wrapped, hijack):
# super().__init__(wrapped, hijack)
# self.tokenizer = wrapped.tokenizer
#
# vocab = self.tokenizer.get_vocab()
#
# self.comma_token = vocab.get(',</w>', None)
#
# self.token_mults = {}
# tokens_with_parens = [(k, v) for k, v in vocab.items() if '(' in k or ')' in k or '[' in k or ']' in k]
# for text, ident in tokens_with_parens:
# mult = 1.0
# for c in text:
# if c == '[':
# mult /= 1.1
# if c == ']':
# mult *= 1.1
# if c == '(':
# mult *= 1.1
# if c == ')':
# mult /= 1.1
#
# if mult != 1.0:
# self.token_mults[ident] = mult
#
# self.id_start = self.wrapped.tokenizer.bos_token_id
# self.id_end = self.wrapped.tokenizer.eos_token_id
# self.id_pad = self.id_end
#
# def tokenize(self, texts):
# tokenized = self.wrapped.tokenizer(texts, truncation=False, add_special_tokens=False)["input_ids"]
#
# return tokenized
#
# def encode_with_transformers(self, tokens):
# outputs = self.wrapped.transformer(input_ids=tokens, output_hidden_states=-opts.CLIP_stop_at_last_layers)
#
# if opts.CLIP_stop_at_last_layers > 1:
# z = outputs.hidden_states[-opts.CLIP_stop_at_last_layers]
# z = self.wrapped.transformer.text_model.final_layer_norm(z)
# else:
# z = outputs.last_hidden_state
#
# return z
#
# def encode_embedding_init_text(self, init_text, nvpt):
# embedding_layer = self.wrapped.transformer.text_model.embeddings
# ids = self.wrapped.tokenizer(init_text, max_length=nvpt, return_tensors="pt", add_special_tokens=False)["input_ids"]
# embedded = embedding_layer.token_embedding.wrapped(ids.to(embedding_layer.token_embedding.wrapped.weight.device)).squeeze(0)
#
# return embedded
#
#
# class FrozenCLIPEmbedderForSDXLWithCustomWords(FrozenCLIPEmbedderWithCustomWords):
# def __init__(self, wrapped, hijack):
# super().__init__(wrapped, hijack)
#
# def encode_with_transformers(self, tokens):
# outputs = self.wrapped.transformer(input_ids=tokens, output_hidden_states=self.wrapped.layer == "hidden")
#
# if opts.sdxl_clip_l_skip is True:
# z = outputs.hidden_states[-opts.CLIP_stop_at_last_layers]
# elif self.wrapped.layer == "last":
# z = outputs.last_hidden_state
# else:
# z = outputs.hidden_states[self.wrapped.layer_idx]
#
# return z
|