Spaces:
Build error
Build error
File size: 21,447 Bytes
699163a |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 |
import ast
import base64
import copy
import io
import json
import logging
import re
from datetime import datetime
from pathlib import Path
import yaml
from PIL import Image
import modules.shared as shared
from modules.extensions import apply_extensions
from modules.html_generator import chat_html_wrapper, make_thumbnail
from modules.text_generation import (generate_reply, get_encoded_length,
get_max_prompt_length)
from modules.utils import replace_all
def get_turn_substrings(state, instruct=False):
if instruct:
if 'turn_template' not in state or state['turn_template'] == '':
template = '<|user|>\n<|user-message|>\n<|bot|>\n<|bot-message|>\n'
else:
template = state['turn_template'].replace(r'\n', '\n')
else:
template = '<|user|>: <|user-message|>\n<|bot|>: <|bot-message|>\n'
replacements = {
'<|user|>': state['name1_instruct' if instruct else 'name1'].strip(),
'<|bot|>': state['name2_instruct' if instruct else 'name2'].strip(),
}
output = {
'user_turn': template.split('<|bot|>')[0],
'bot_turn': '<|bot|>' + template.split('<|bot|>')[1],
'user_turn_stripped': template.split('<|bot|>')[0].split('<|user-message|>')[0],
'bot_turn_stripped': '<|bot|>' + template.split('<|bot|>')[1].split('<|bot-message|>')[0],
}
for k in output:
output[k] = replace_all(output[k], replacements)
return output
def generate_chat_prompt(user_input, state, **kwargs):
impersonate = kwargs.get('impersonate', False)
_continue = kwargs.get('_continue', False)
also_return_rows = kwargs.get('also_return_rows', False)
history = state.get('history', shared.history['internal'])
is_instruct = state['mode'] == 'instruct'
# Finding the maximum prompt size
chat_prompt_size = state['chat_prompt_size']
if shared.soft_prompt:
chat_prompt_size -= shared.soft_prompt_tensor.shape[1]
max_length = min(get_max_prompt_length(state), chat_prompt_size)
all_substrings = {
'chat': get_turn_substrings(state, instruct=False),
'instruct': get_turn_substrings(state, instruct=True)
}
substrings = all_substrings['instruct' if is_instruct else 'chat']
# Creating the template for "chat-instruct" mode
if state['mode'] == 'chat-instruct':
wrapper = ''
command = state['chat-instruct_command'].replace('<|character|>', state['name2'] if not impersonate else state['name1'])
wrapper += state['context_instruct']
wrapper += all_substrings['instruct']['user_turn'].replace('<|user-message|>', command)
wrapper += all_substrings['instruct']['bot_turn_stripped']
if impersonate:
wrapper += substrings['user_turn_stripped'].rstrip(' ')
else:
wrapper += apply_extensions("bot_prefix", substrings['bot_turn_stripped'].rstrip(' '))
else:
wrapper = '<|prompt|>'
# Building the prompt
min_rows = 3
i = len(history) - 1
rows = [state['context_instruct'] if is_instruct else f"{state['context'].strip()}\n"]
while i >= 0 and get_encoded_length(wrapper.replace('<|prompt|>', ''.join(rows))) < max_length:
if _continue and i == len(history) - 1:
rows.insert(1, substrings['bot_turn_stripped'] + history[i][1].strip())
else:
rows.insert(1, substrings['bot_turn'].replace('<|bot-message|>', history[i][1].strip()))
string = history[i][0]
if string not in ['', '<|BEGIN-VISIBLE-CHAT|>']:
rows.insert(1, replace_all(substrings['user_turn'], {'<|user-message|>': string.strip(), '<|round|>': str(i)}))
i -= 1
if impersonate:
if state['mode'] == 'chat-instruct':
min_rows = 1
else:
min_rows = 2
rows.append(substrings['user_turn_stripped'].rstrip(' '))
elif not _continue:
# Adding the user message
if len(user_input) > 0:
rows.append(replace_all(substrings['user_turn'], {'<|user-message|>': user_input.strip(), '<|round|>': str(len(history))}))
# Adding the Character prefix
if state['mode'] != 'chat-instruct':
rows.append(apply_extensions("bot_prefix", substrings['bot_turn_stripped'].rstrip(' ')))
while len(rows) > min_rows and get_encoded_length(wrapper.replace('<|prompt|>', ''.join(rows))) >= max_length:
rows.pop(1)
prompt = wrapper.replace('<|prompt|>', ''.join(rows))
if also_return_rows:
return prompt, rows
else:
return prompt
def get_stopping_strings(state):
stopping_strings = []
if state['mode'] in ['instruct', 'chat-instruct']:
stopping_strings += [
state['turn_template'].split('<|user-message|>')[1].split('<|bot|>')[0] + '<|bot|>',
state['turn_template'].split('<|bot-message|>')[1] + '<|user|>'
]
replacements = {
'<|user|>': state['name1_instruct'],
'<|bot|>': state['name2_instruct']
}
for i in range(len(stopping_strings)):
stopping_strings[i] = replace_all(stopping_strings[i], replacements).rstrip(' ').replace(r'\n', '\n')
if state['mode'] in ['chat', 'chat-instruct']:
stopping_strings += [
f"\n{state['name1']}:",
f"\n{state['name2']}:"
]
stopping_strings += ast.literal_eval(f"[{state['custom_stopping_strings']}]")
return stopping_strings
def extract_message_from_reply(reply, state):
next_character_found = False
stopping_strings = get_stopping_strings(state)
if state['stop_at_newline']:
lines = reply.split('\n')
reply = lines[0].strip()
if len(lines) > 1:
next_character_found = True
else:
for string in stopping_strings:
idx = reply.find(string)
if idx != -1:
reply = reply[:idx]
next_character_found = True
# If something like "\nYo" is generated just before "\nYou:"
# is completed, trim it
if not next_character_found:
for string in stopping_strings:
for j in range(len(string) - 1, 0, -1):
if reply[-j:] == string[:j]:
reply = reply[:-j]
break
else:
continue
break
return reply, next_character_found
def chatbot_wrapper(text, state, regenerate=False, _continue=False):
if shared.model_name == 'None' or shared.model is None:
logging.error("No model is loaded! Select one in the Model tab.")
yield shared.history['visible']
return
# Defining some variables
cumulative_reply = ''
just_started = True
visible_text = None
eos_token = '\n' if state['stop_at_newline'] else None
stopping_strings = get_stopping_strings(state)
# Preparing the input
if not any((regenerate, _continue)):
text, visible_text = apply_extensions('input_hijack', text, visible_text)
if visible_text is None:
visible_text = text
text = apply_extensions('input', text)
# *Is typing...*
yield shared.history['visible'] + [[visible_text, shared.processing_message]]
else:
text, visible_text = shared.history['internal'][-1][0], shared.history['visible'][-1][0]
if regenerate:
shared.history['visible'].pop()
shared.history['internal'].pop()
# *Is typing...*
yield shared.history['visible'] + [[visible_text, shared.processing_message]]
elif _continue:
last_reply = [shared.history['internal'][-1][1], shared.history['visible'][-1][1]]
yield shared.history['visible'][:-1] + [[visible_text, last_reply[1] + '...']]
# Generating the prompt
kwargs = {'_continue': _continue}
prompt = apply_extensions('custom_generate_chat_prompt', text, state, **kwargs)
if prompt is None:
prompt = generate_chat_prompt(text, state, **kwargs)
# Generate
for i in range(state['chat_generation_attempts']):
reply = None
for j, reply in enumerate(generate_reply(prompt + cumulative_reply, state, eos_token=eos_token, stopping_strings=stopping_strings, is_chat=True)):
reply = cumulative_reply + reply
# Extracting the reply
reply, next_character_found = extract_message_from_reply(reply, state)
visible_reply = re.sub("(<USER>|<user>|{{user}})", state['name1'], reply)
visible_reply = apply_extensions("output", visible_reply)
# We need this global variable to handle the Stop event,
# otherwise gradio gets confused
if shared.stop_everything:
return shared.history['visible']
if just_started:
just_started = False
if not _continue:
shared.history['internal'].append(['', ''])
shared.history['visible'].append(['', ''])
if _continue:
shared.history['internal'][-1] = [text, last_reply[0] + reply]
shared.history['visible'][-1] = [visible_text, last_reply[1] + visible_reply]
yield shared.history['visible']
elif not (j == 0 and visible_reply.strip() == ''):
shared.history['internal'][-1] = [text, reply]
shared.history['visible'][-1] = [visible_text, visible_reply]
yield shared.history['visible']
if next_character_found:
break
if reply in [None, '']:
break
else:
cumulative_reply = reply
yield shared.history['visible']
def impersonate_wrapper(text, state):
if shared.model_name == 'None' or shared.model is None:
logging.error("No model is loaded! Select one in the Model tab.")
yield ''
return
# Defining some variables
cumulative_reply = ''
eos_token = '\n' if state['stop_at_newline'] else None
prompt = generate_chat_prompt('', state, impersonate=True)
stopping_strings = get_stopping_strings(state)
yield text + '...'
cumulative_reply = text
for i in range(state['chat_generation_attempts']):
reply = None
for reply in generate_reply(prompt + cumulative_reply, state, eos_token=eos_token, stopping_strings=stopping_strings, is_chat=True):
reply = cumulative_reply + reply
reply, next_character_found = extract_message_from_reply(reply, state)
yield reply
if next_character_found:
break
if reply in [None, '']:
break
else:
cumulative_reply = reply
yield cumulative_reply
def generate_chat_reply(text, state, regenerate=False, _continue=False):
if regenerate or _continue:
text = ''
if (len(shared.history['visible']) == 1 and not shared.history['visible'][0][0]) or len(shared.history['internal']) == 0:
yield shared.history['visible']
return
for history in chatbot_wrapper(text, state, regenerate=regenerate, _continue=_continue):
yield history
# Same as above but returns HTML
def generate_chat_reply_wrapper(text, state, regenerate=False, _continue=False):
for history in generate_chat_reply(text, state, regenerate, _continue):
yield chat_html_wrapper(history, state['name1'], state['name2'], state['mode'], state['chat_style'])
def remove_last_message():
if len(shared.history['visible']) > 0 and shared.history['internal'][-1][0] != '<|BEGIN-VISIBLE-CHAT|>':
last = shared.history['visible'].pop()
shared.history['internal'].pop()
else:
last = ['', '']
return last[0]
def send_last_reply_to_input():
if len(shared.history['internal']) > 0:
return shared.history['internal'][-1][1]
else:
return ''
def replace_last_reply(text):
if len(shared.history['visible']) > 0:
shared.history['visible'][-1][1] = text
shared.history['internal'][-1][1] = apply_extensions("input", text)
def send_dummy_message(text):
shared.history['visible'].append([text, ''])
shared.history['internal'].append([apply_extensions("input", text), ''])
def send_dummy_reply(text):
if len(shared.history['visible']) > 0 and not shared.history['visible'][-1][1] == '':
shared.history['visible'].append(['', ''])
shared.history['internal'].append(['', ''])
shared.history['visible'][-1][1] = text
shared.history['internal'][-1][1] = apply_extensions("input", text)
def clear_chat_log(greeting, mode):
shared.history['visible'] = []
shared.history['internal'] = []
if mode != 'instruct':
if greeting != '':
shared.history['internal'] += [['<|BEGIN-VISIBLE-CHAT|>', greeting]]
shared.history['visible'] += [['', apply_extensions("output", greeting)]]
save_history(mode)
def redraw_html(name1, name2, mode, style, reset_cache=False):
return chat_html_wrapper(shared.history['visible'], name1, name2, mode, style, reset_cache=reset_cache)
def tokenize_dialogue(dialogue, name1, name2):
history = []
messages = []
dialogue = re.sub('<START>', '', dialogue)
dialogue = re.sub('<start>', '', dialogue)
dialogue = re.sub('(\n|^)[Aa]non:', '\\1You:', dialogue)
dialogue = re.sub('(\n|^)\[CHARACTER\]:', f'\\g<1>{name2}:', dialogue)
idx = [m.start() for m in re.finditer(f"(^|\n)({re.escape(name1)}|{re.escape(name2)}):", dialogue)]
if len(idx) == 0:
return history
for i in range(len(idx) - 1):
messages.append(dialogue[idx[i]:idx[i + 1]].strip())
messages.append(dialogue[idx[-1]:].strip())
entry = ['', '']
for i in messages:
if i.startswith(f'{name1}:'):
entry[0] = i[len(f'{name1}:'):].strip()
elif i.startswith(f'{name2}:'):
entry[1] = i[len(f'{name2}:'):].strip()
if not (len(entry[0]) == 0 and len(entry[1]) == 0):
history.append(entry)
entry = ['', '']
print("\033[1;32;1m\nDialogue tokenized to:\033[0;37;0m\n", end='')
for row in history:
for column in row:
print("\n")
for line in column.strip().split('\n'):
print("| " + line + "\n")
print("|\n")
print("------------------------------")
return history
def save_history(mode, timestamp=False):
# Instruct mode histories should not be saved as if
# Alpaca or Vicuna were characters
if mode == 'instruct':
if not timestamp:
return
fname = f"Instruct_{datetime.now().strftime('%Y%m%d-%H%M%S')}.json"
else:
if timestamp:
fname = f"{shared.character}_{datetime.now().strftime('%Y%m%d-%H%M%S')}.json"
else:
fname = f"{shared.character}_persistent.json"
if not Path('logs').exists():
Path('logs').mkdir()
with open(Path(f'logs/{fname}'), 'w', encoding='utf-8') as f:
f.write(json.dumps({'data': shared.history['internal'], 'data_visible': shared.history['visible']}, indent=2))
return Path(f'logs/{fname}')
def load_history(file, name1, name2):
file = file.decode('utf-8')
try:
j = json.loads(file)
if 'data' in j:
shared.history['internal'] = j['data']
if 'data_visible' in j:
shared.history['visible'] = j['data_visible']
else:
shared.history['visible'] = copy.deepcopy(shared.history['internal'])
except:
shared.history['internal'] = tokenize_dialogue(file, name1, name2)
shared.history['visible'] = copy.deepcopy(shared.history['internal'])
def replace_character_names(text, name1, name2):
text = text.replace('{{user}}', name1).replace('{{char}}', name2)
return text.replace('<USER>', name1).replace('<BOT>', name2)
def build_pygmalion_style_context(data):
context = ""
if 'char_persona' in data and data['char_persona'] != '':
context += f"{data['char_name']}'s Persona: {data['char_persona']}\n"
if 'world_scenario' in data and data['world_scenario'] != '':
context += f"Scenario: {data['world_scenario']}\n"
context = f"{context.strip()}\n<START>\n"
return context
def generate_pfp_cache(character):
cache_folder = Path("cache")
if not cache_folder.exists():
cache_folder.mkdir()
for path in [Path(f"characters/{character}.{extension}") for extension in ['png', 'jpg', 'jpeg']]:
if path.exists():
img = make_thumbnail(Image.open(path))
img.save(Path('cache/pfp_character.png'), format='PNG')
return img
return None
def load_character(character, name1, name2, instruct=False):
shared.character = character
context = greeting = turn_template = ""
greeting_field = 'greeting'
picture = None
# Deleting the profile picture cache, if any
if Path("cache/pfp_character.png").exists():
Path("cache/pfp_character.png").unlink()
if character != 'None':
folder = 'characters' if not instruct else 'characters/instruction-following'
picture = generate_pfp_cache(character)
for extension in ["yml", "yaml", "json"]:
filepath = Path(f'{folder}/{character}.{extension}')
if filepath.exists():
break
file_contents = open(filepath, 'r', encoding='utf-8').read()
data = json.loads(file_contents) if extension == "json" else yaml.safe_load(file_contents)
# Finding the bot's name
for k in ['name', 'bot', '<|bot|>', 'char_name']:
if k in data and data[k] != '':
name2 = data[k]
break
# Find the user name (if any)
for k in ['your_name', 'user', '<|user|>']:
if k in data and data[k] != '':
name1 = data[k]
break
for field in ['context', 'greeting', 'example_dialogue', 'char_persona', 'char_greeting', 'world_scenario']:
if field in data:
data[field] = replace_character_names(data[field], name1, name2)
if 'context' in data:
context = data['context']
if not instruct:
context = context.strip() + '\n'
elif "char_persona" in data:
context = build_pygmalion_style_context(data)
greeting_field = 'char_greeting'
if 'example_dialogue' in data:
context += f"{data['example_dialogue'].strip()}\n"
if greeting_field in data:
greeting = data[greeting_field]
if 'turn_template' in data:
turn_template = data['turn_template']
else:
context = shared.settings['context']
name2 = shared.settings['name2']
greeting = shared.settings['greeting']
turn_template = shared.settings['turn_template']
if not instruct:
shared.history['internal'] = []
shared.history['visible'] = []
if Path(f'logs/{shared.character}_persistent.json').exists():
load_history(open(Path(f'logs/{shared.character}_persistent.json'), 'rb').read(), name1, name2)
else:
# Insert greeting if it exists
if greeting != "":
shared.history['internal'] += [['<|BEGIN-VISIBLE-CHAT|>', greeting]]
shared.history['visible'] += [['', apply_extensions("output", greeting)]]
# Create .json log files since they don't already exist
save_history('instruct' if instruct else 'chat')
return name1, name2, picture, greeting, context, repr(turn_template)[1:-1]
def upload_character(json_file, img, tavern=False):
json_file = json_file if type(json_file) == str else json_file.decode('utf-8')
data = json.loads(json_file)
outfile_name = data["char_name"]
i = 1
while Path(f'characters/{outfile_name}.json').exists():
outfile_name = f'{data["char_name"]}_{i:03d}'
i += 1
if tavern:
outfile_name = f'TavernAI-{outfile_name}'
with open(Path(f'characters/{outfile_name}.json'), 'w', encoding='utf-8') as f:
f.write(json_file)
if img is not None:
img = Image.open(io.BytesIO(img))
img.save(Path(f'characters/{outfile_name}.png'))
logging.info(f'New character saved to "characters/{outfile_name}.json".')
return outfile_name
def upload_tavern_character(img, name1, name2):
_img = Image.open(io.BytesIO(img))
_img.getexif()
decoded_string = base64.b64decode(_img.info['chara'])
_json = json.loads(decoded_string)
_json = {"char_name": _json['name'], "char_persona": _json['description'], "char_greeting": _json["first_mes"], "example_dialogue": _json['mes_example'], "world_scenario": _json['scenario']}
return upload_character(json.dumps(_json), img, tavern=True)
def upload_your_profile_picture(img):
cache_folder = Path("cache")
if not cache_folder.exists():
cache_folder.mkdir()
if img is None:
if Path("cache/pfp_me.png").exists():
Path("cache/pfp_me.png").unlink()
else:
img = make_thumbnail(img)
img.save(Path('cache/pfp_me.png'))
logging.info('Profile picture saved to "cache/pfp_me.png"')
|