431 lines
No EOL
14 KiB
Python
431 lines
No EOL
14 KiB
Python
import json
|
||
import os
|
||
from pathlib import Path
|
||
import re
|
||
import sys
|
||
import textwrap
|
||
import threading
|
||
import time
|
||
import traceback
|
||
import tiktoken
|
||
|
||
from colorama import Fore
|
||
from dotenv import load_dotenv
|
||
import openai
|
||
from retry import retry
|
||
from tqdm import tqdm
|
||
|
||
#Globals
|
||
load_dotenv()
|
||
if os.getenv('api').replace(' ', '') != '':
|
||
openai.api_base = os.getenv('api')
|
||
|
||
openai.organization = os.getenv('org')
|
||
openai.api_key = os.getenv('key')
|
||
MODEL = os.getenv('model')
|
||
TIMEOUT = int(os.getenv('timeout'))
|
||
LANGUAGE=os.getenv('language').capitalize()
|
||
|
||
INPUTAPICOST = .002 # Depends on the model https://openai.com/pricing
|
||
OUTPUTAPICOST = .002
|
||
PROMPT = Path('prompt.txt').read_text(encoding='utf-8')
|
||
THREADS = int(os.getenv('threads')) # Controls how many threads are working on a single file (May have to drop this)
|
||
LOCK = threading.Lock()
|
||
WIDTH = int(os.getenv('width'))
|
||
LISTWIDTH = int(os.getenv('listWidth'))
|
||
NOTEWIDTH = 50
|
||
MAXHISTORY = 10
|
||
ESTIMATE = ''
|
||
totalTokens = [0, 0]
|
||
NAMESLIST = []
|
||
|
||
#tqdm Globals
|
||
BAR_FORMAT='{l_bar}{bar:10}{r_bar}{bar:-10b}'
|
||
POSITION=0
|
||
LEAVE=False
|
||
BRFLAG = False # If the game uses <br> instead
|
||
FIXTEXTWRAP = True
|
||
IGNORETLTEXT = True
|
||
|
||
def handleAnim(filename, estimate):
|
||
global ESTIMATE, totalTokens
|
||
ESTIMATE = estimate
|
||
|
||
if estimate:
|
||
start = time.time()
|
||
translatedData = openFiles(filename)
|
||
|
||
# Print Result
|
||
end = time.time()
|
||
tqdm.write(getResultString(translatedData, end - start, filename))
|
||
with LOCK:
|
||
totalTokens[0] += translatedData[1][0]
|
||
totalTokens[1] += translatedData[1][1]
|
||
|
||
return getResultString(['', totalTokens, None], end - start, 'TOTAL')
|
||
|
||
else:
|
||
try:
|
||
with open('translated/' + filename, 'w', encoding='UTF-8') as outFile:
|
||
start = time.time()
|
||
translatedData = openFiles(filename)
|
||
|
||
# Print Result
|
||
end = time.time()
|
||
json.dump(translatedData[0], outFile, ensure_ascii=False)
|
||
tqdm.write(getResultString(translatedData, end - start, filename))
|
||
with LOCK:
|
||
totalTokens[0] += translatedData[1][0]
|
||
totalTokens[1] += translatedData[1][1]
|
||
except Exception as e:
|
||
return 'Fail'
|
||
|
||
return getResultString(['', totalTokens, None], end - start, 'TOTAL')
|
||
|
||
def openFiles(filename):
|
||
with open('files/' + filename, 'r', encoding='UTF-8-sig') as f:
|
||
data = json.load(f)
|
||
|
||
# Map Files
|
||
if '.json' in filename:
|
||
translatedData = parseJSON(data, filename)
|
||
|
||
else:
|
||
raise NameError(filename + ' Not Supported')
|
||
|
||
return translatedData
|
||
|
||
def getResultString(translatedData, translationTime, filename):
|
||
# File Print String
|
||
totalTokenstring =\
|
||
Fore.YELLOW +\
|
||
'[Input: ' + str(translatedData[1][0]) + ']'\
|
||
'[Output: ' + str(translatedData[1][1]) + ']'\
|
||
'[Cost: ${:,.4f}'.format((translatedData[1][0] * .001 * INPUTAPICOST) +\
|
||
(translatedData[1][1] * .001 * OUTPUTAPICOST)) + ']'
|
||
timeString = Fore.BLUE + '[' + str(round(translationTime, 1)) + 's]'
|
||
|
||
if translatedData[2] == None:
|
||
# Success
|
||
return filename + ': ' + totalTokenstring + timeString + Fore.GREEN + u' \u2713 ' + Fore.RESET
|
||
|
||
else:
|
||
# Fail
|
||
try:
|
||
raise translatedData[2]
|
||
except Exception as e:
|
||
traceback.print_exc()
|
||
errorString = str(e) + Fore.RED
|
||
return filename + ': ' + totalTokenstring + timeString + Fore.RED + u' \u2717 ' +\
|
||
errorString + Fore.RESET
|
||
|
||
def parseJSON(data, filename):
|
||
totalTokens = [0, 0]
|
||
totalLines = 0
|
||
totalLines = len(data)
|
||
global LOCK
|
||
|
||
with tqdm(bar_format=BAR_FORMAT, position=POSITION, total=totalLines, leave=LEAVE) as pbar:
|
||
pbar.desc=filename
|
||
pbar.total=totalLines
|
||
try:
|
||
result = translateJSON(data, pbar)
|
||
totalTokens[0] += result[0]
|
||
totalTokens[1] += result[1]
|
||
except Exception as e:
|
||
return [data, totalTokens, e]
|
||
return [data, totalTokens, None]
|
||
|
||
def translateJSON(data, pbar):
|
||
textHistory = []
|
||
maxHistory = MAXHISTORY
|
||
tokens = [0, 0]
|
||
|
||
for key, value in data.items():
|
||
# Text
|
||
if value == "":
|
||
jaString = key
|
||
else:
|
||
jaString = value
|
||
|
||
# Check if TLed
|
||
# If there isn't any Japanese in the text just skip
|
||
if IGNORETLTEXT is True:
|
||
if not re.search(r'[一-龠]+|[ぁ-ゔ]+|[ァ-ヴー]+', jaString):
|
||
pbar.update(1)
|
||
continue
|
||
|
||
# Remove any textwrap
|
||
if FIXTEXTWRAP == True:
|
||
jaString = jaString.replace('\n', ' ')
|
||
|
||
# Translate
|
||
if jaString != '':
|
||
response = translateGPT(f'{jaString}', textHistory, True)
|
||
tokens[0] += response[1][0]
|
||
tokens[1] += response[1][1]
|
||
translatedText = response[0]
|
||
textHistory.append('\"' + translatedText + '\"')
|
||
else:
|
||
translatedText = jaString
|
||
textHistory.append('\"' + translatedText + '\"')
|
||
|
||
# Remove added speaker
|
||
translatedText = re.sub(r'^.+?\s\|\s?', '', translatedText)
|
||
|
||
# Textwrap
|
||
if '\n' not in translatedText:
|
||
translatedText = textwrap.fill(translatedText, width=WIDTH)
|
||
|
||
# Set Data
|
||
data[key] = translatedText
|
||
|
||
# Keep textHistory list at length maxHistory
|
||
if len(textHistory) > maxHistory:
|
||
textHistory.pop(0)
|
||
currentGroup = []
|
||
pbar.update(1)
|
||
|
||
return tokens
|
||
|
||
def subVars(jaString):
|
||
jaString = jaString.replace('\u3000', ' ')
|
||
|
||
# Nested
|
||
count = 0
|
||
nestedList = re.findall(r'[\\]+[\w]+\[[\\]+[\w]+\[[0-9]+\]\]', jaString)
|
||
nestedList = set(nestedList)
|
||
if len(nestedList) != 0:
|
||
for icon in nestedList:
|
||
jaString = jaString.replace(icon, '{Nested_' + str(count) + '}')
|
||
count += 1
|
||
|
||
# Icons
|
||
count = 0
|
||
iconList = re.findall(r'[\\]+[iIkKwWaA]+\[[0-9]+\]', jaString)
|
||
iconList = set(iconList)
|
||
if len(iconList) != 0:
|
||
for icon in iconList:
|
||
jaString = jaString.replace(icon, '{Ascii_' + str(count) + '}')
|
||
count += 1
|
||
|
||
# Colors
|
||
count = 0
|
||
colorList = re.findall(r'[\\]+[cC]\[[0-9]+\]', jaString)
|
||
colorList = set(colorList)
|
||
if len(colorList) != 0:
|
||
for color in colorList:
|
||
jaString = jaString.replace(color, '{Color_' + str(count) + '}')
|
||
count += 1
|
||
|
||
# Names
|
||
count = 0
|
||
nameList = re.findall(r'[\\]+[nN]\[.+?\]+', jaString)
|
||
nameList = set(nameList)
|
||
if len(nameList) != 0:
|
||
for name in nameList:
|
||
jaString = jaString.replace(name, '{Noun_' + str(count) + '}')
|
||
count += 1
|
||
|
||
# Variables
|
||
count = 0
|
||
varList = re.findall(r'[\\]+[vV]\[[0-9]+\]', jaString)
|
||
varList = set(varList)
|
||
if len(varList) != 0:
|
||
for var in varList:
|
||
jaString = jaString.replace(var, '{Var_' + str(count) + '}')
|
||
count += 1
|
||
|
||
# Formatting
|
||
count = 0
|
||
formatList = re.findall(r'[\\]+[\w]+\[.+?\]', jaString)
|
||
formatList = set(formatList)
|
||
if len(formatList) != 0:
|
||
for var in formatList:
|
||
jaString = jaString.replace(var, '{FCode_' + str(count) + '}')
|
||
count += 1
|
||
|
||
# Put all lists in list and return
|
||
allList = [nestedList, iconList, colorList, nameList, varList, formatList]
|
||
return [jaString, allList]
|
||
|
||
def resubVars(translatedText, allList):
|
||
# Fix Spacing and ChatGPT Nonsense
|
||
matchList = re.findall(r'\[\s?.+?\s?\]', translatedText)
|
||
if len(matchList) > 0:
|
||
for match in matchList:
|
||
text = match.strip()
|
||
translatedText = translatedText.replace(match, text)
|
||
|
||
# Nested
|
||
count = 0
|
||
if len(allList[0]) != 0:
|
||
for var in allList[0]:
|
||
translatedText = translatedText.replace('{Nested_' + str(count) + '}', var)
|
||
count += 1
|
||
|
||
# Icons
|
||
count = 0
|
||
if len(allList[1]) != 0:
|
||
for var in allList[1]:
|
||
translatedText = translatedText.replace('{Ascii_' + str(count) + '}', var)
|
||
count += 1
|
||
|
||
# Colors
|
||
count = 0
|
||
if len(allList[2]) != 0:
|
||
for var in allList[2]:
|
||
translatedText = translatedText.replace('{Color_' + str(count) + '}', var)
|
||
count += 1
|
||
|
||
# Names
|
||
count = 0
|
||
if len(allList[3]) != 0:
|
||
for var in allList[3]:
|
||
translatedText = translatedText.replace('{Noun_' + str(count) + '}', var)
|
||
count += 1
|
||
|
||
# Vars
|
||
count = 0
|
||
if len(allList[4]) != 0:
|
||
for var in allList[4]:
|
||
translatedText = translatedText.replace('{Var_' + str(count) + '}', var)
|
||
count += 1
|
||
|
||
# Formatting
|
||
count = 0
|
||
if len(allList[5]) != 0:
|
||
for var in allList[5]:
|
||
translatedText = translatedText.replace('{FCode_' + str(count) + '}', var)
|
||
count += 1
|
||
|
||
return translatedText
|
||
|
||
def batchList(input_list, batch_size):
|
||
if not isinstance(batch_size, int) or batch_size <= 0:
|
||
raise ValueError("batch_size must be a positive integer")
|
||
|
||
return [input_list[i:i + batch_size] for i in range(0, len(input_list), batch_size)]
|
||
|
||
def createContext(fullPromptFlag, subbedT):
|
||
characters = 'Game Characters:\
|
||
ボク == Boku - Male\
|
||
ユイ == Yui - Female\
|
||
ヒロミ == Hiromi - Female\
|
||
ミヤビ == Miyabi - Female\
|
||
ショウコ == Shoko - Female\
|
||
リリ == Riri - Female\
|
||
ララ == Rara - Female\
|
||
ミユキ == Miyuki - Female\
|
||
ナギサ == Nagisa - Female\
|
||
タケル == Takeru - Male'
|
||
system = PROMPT if fullPromptFlag else \
|
||
f'Output ONLY the {LANGUAGE} translation in the following format: `Translation: <{LANGUAGE.upper()}_TRANSLATION>`'
|
||
user = f'Line to Translate = {subbedT}'
|
||
return characters, system, user
|
||
|
||
def translateText(subbedT, history, fullPromptFlag):
|
||
characters, system, user = createContext(fullPromptFlag, subbedT)
|
||
# Prompt
|
||
msg = [{"role": "system", "content": system}]
|
||
|
||
# Characters
|
||
msg.append({"role": "user", "content": characters})
|
||
|
||
# History
|
||
if isinstance(history, list):
|
||
msg.extend([{"role": "user", "content": h} for h in history])
|
||
else:
|
||
msg.append({"role": "user", "content": history})
|
||
|
||
# Content to TL
|
||
msg.append({"role": "user", "content": user})
|
||
response = openai.ChatCompletion.create(
|
||
temperature=0,
|
||
frequency_penalty=0,
|
||
presence_penalty=0,
|
||
model=MODEL,
|
||
messages=msg,
|
||
request_timeout=TIMEOUT,
|
||
)
|
||
return response
|
||
|
||
def cleanTranslatedText(translatedText, varResponse):
|
||
placeholders = {
|
||
f'{LANGUAGE} Translation: ': '',
|
||
'Translation: ': '',
|
||
# Add more replacements as needed
|
||
}
|
||
for target, replacement in placeholders.items():
|
||
translatedText = translatedText.replace(target, replacement)
|
||
|
||
translatedText = resubVars(translatedText, varResponse[1])
|
||
return [line for line in translatedText.split('\n') if line]
|
||
|
||
def extractTranslation(translatedTextList, is_list):
|
||
pattern = r'L(\d+) - (.*)'
|
||
# If it's a batch (i.e., list), extract with tags; otherwise, return the single item.
|
||
if is_list:
|
||
return [re.findall(pattern, line)[0][1] for line in translatedTextList if re.search(pattern, line)]
|
||
else:
|
||
matchList = re.findall(pattern, translatedTextList)
|
||
return matchList[0][1] if matchList else translatedTextList
|
||
|
||
def countTokens(tItem, history):
|
||
enc = tiktoken.encoding_for_model(MODEL)
|
||
encode_count = lambda item: sum(len(enc.encode(line)) for line in (item if isinstance(item, list) else [item]))
|
||
inputTotalTokens = encode_count(history) + encode_count(PROMPT)
|
||
outputTotalTokens = encode_count(tItem) * 2 # Estimated
|
||
return inputTotalTokens + outputTotalTokens
|
||
|
||
def combineList(tlist, text):
|
||
if isinstance(text, list):
|
||
return [t for sublist in tlist for t in sublist]
|
||
return tlist[0]
|
||
|
||
@retry(exceptions=Exception, tries=5, delay=5)
|
||
def translateGPT(text, history, fullPromptFlag):
|
||
totalTokens = [0, 0]
|
||
if isinstance(text, list):
|
||
tList = batchList(text, BATCHSIZE)
|
||
history = ''
|
||
else:
|
||
tList = [text]
|
||
|
||
for index, tItem in enumerate(tList):
|
||
# Before sending to translation, if we have a list of items, add the formatting
|
||
if isinstance(tItem, list):
|
||
payload = '\n'.join([f'L{i} - {item}' for i, item in enumerate(tItem)])
|
||
varResponse = subVars(payload)
|
||
subbedT = varResponse[0]
|
||
else:
|
||
varResponse = subVars(tItem)
|
||
subbedT = varResponse[0]
|
||
|
||
# Things to Check before starting translation
|
||
if not re.search(r'[一-龠ぁ-ゔァ-ヴーａ-ｚＡ-Ｚ０-９]+', subbedT):
|
||
continue
|
||
|
||
if ESTIMATE:
|
||
totalTokens[0] += countTokens(tItem, history)
|
||
continue
|
||
|
||
# Translating
|
||
response = translateText(subbedT, history, fullPromptFlag)
|
||
translatedText = response.choices[0].message.content
|
||
totalTokens[0] += response.usage.prompt_tokens
|
||
totalTokens[1] += response.usage.completion_tokens
|
||
|
||
# Formatting
|
||
translatedTextList = cleanTranslatedText(translatedText, varResponse)
|
||
if isinstance(tItem, list):
|
||
extractedTranslations = extractTranslation(translatedTextList, True)
|
||
tList[index] = extractedTranslations
|
||
history = extractedTranslations[-10:] # Update history if we have a list
|
||
else:
|
||
# Ensure we're passing a single string to extractTranslation
|
||
extractedTranslations = extractTranslation('\n'.join(translatedTextList), False)
|
||
tList[index] = extractedTranslations
|
||
|
||
finalList = combineList(tList, text)
|
||
return [finalList, totalTokens] |