379 lines
No EOL
12 KiB
Python
379 lines
No EOL
12 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()
|
|
openai.organization = os.getenv('org')
|
|
openai.api_key = os.getenv('key')
|
|
|
|
INPUTAPICOST = .002 # Depends on the model https://openai.com/pricing
|
|
OUTPUTAPICOST = .002
|
|
PROMPT = Path('prompt.txt').read_text(encoding='utf-8')
|
|
THREADS = 10 # Controls how many threads are working on a single file (May have to drop this)
|
|
LOCK = threading.Lock()
|
|
WIDTH = 50
|
|
LISTWIDTH = 90
|
|
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 = False
|
|
|
|
def handleJSON(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]
|
|
speaker = 'None'
|
|
|
|
for item in data.items():
|
|
# Speaker
|
|
if 'name' in item[1]:
|
|
if item[1]['name'] not in [None, '-']:
|
|
response = translateGPT(item[1]['name'], 'Reply with only the english translation of the NPC name', False)
|
|
speaker = response[0]
|
|
tokens[0] += response[1][0]
|
|
tokens[1] += response[1][1]
|
|
item[1]['name'] = speaker
|
|
else:
|
|
speaker = 'None'
|
|
|
|
# Text
|
|
for text in ['text', 'text2', 'help1', 'help2', 'help3', 'like', 'message']:
|
|
if text in item[1]:
|
|
if item[1][text] != None:
|
|
jaString = item[1][text]
|
|
|
|
# Remove any textwrap
|
|
if FIXTEXTWRAP == True:
|
|
jaString = jaString.replace('\n', ' ')
|
|
|
|
# Translate
|
|
if jaString != '':
|
|
response = translateGPT(f'{speaker} | {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
|
|
translatedText = textwrap.fill(translatedText, width=WIDTH)
|
|
|
|
# Set Data
|
|
item[1][text] = 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', ' ')
|
|
|
|
# Icons
|
|
count = 0
|
|
iconList = re.findall(r'[\\]+[iIkKwW]+\[[0-9]+\]', 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, '[N_' + 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
|
|
if '笑えるよね.' in jaString:
|
|
print('t')
|
|
formatList = re.findall(r'[\\]+CL', 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 = [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)
|
|
|
|
# Icons
|
|
count = 0
|
|
if len(allList[0]) != 0:
|
|
for var in allList[0]:
|
|
translatedText = translatedText.replace('[Ascii_' + str(count) + ']', var)
|
|
count += 1
|
|
|
|
# Colors
|
|
count = 0
|
|
if len(allList[1]) != 0:
|
|
for var in allList[1]:
|
|
translatedText = translatedText.replace('[Color_' + str(count) + ']', var)
|
|
count += 1
|
|
|
|
# Names
|
|
count = 0
|
|
if len(allList[2]) != 0:
|
|
for var in allList[2]:
|
|
translatedText = translatedText.replace('[N_' + str(count) + ']', var)
|
|
count += 1
|
|
|
|
# Vars
|
|
count = 0
|
|
if len(allList[3]) != 0:
|
|
for var in allList[3]:
|
|
translatedText = translatedText.replace('[Var_' + str(count) + ']', var)
|
|
count += 1
|
|
|
|
# Formatting
|
|
count = 0
|
|
if len(allList[4]) != 0:
|
|
for var in allList[4]:
|
|
translatedText = translatedText.replace('[FCode_' + str(count) + ']', var)
|
|
count += 1
|
|
|
|
# Remove Color Variables Spaces
|
|
# if '\\c' in translatedText:
|
|
# translatedText = re.sub(r'\s*(\\+c\[[1-9]+\])\s*', r' \1', translatedText)
|
|
# translatedText = re.sub(r'\s*(\\+c\[0+\])', r'\1', translatedText)
|
|
return translatedText
|
|
|
|
@retry(exceptions=Exception, tries=5, delay=5)
|
|
def translateGPT(t, history, fullPromptFlag):
|
|
# If ESTIMATE is True just count this as an execution and return.
|
|
if ESTIMATE:
|
|
enc = tiktoken.encoding_for_model("gpt-3.5-turbo")
|
|
historyRaw = ''
|
|
if isinstance(history, list):
|
|
for line in history:
|
|
historyRaw += line
|
|
else:
|
|
historyRaw = history
|
|
|
|
inputTotalTokens = len(enc.encode(historyRaw)) + len(enc.encode(PROMPT))
|
|
outputTotalTokens = len(enc.encode(t)) * 2 # Estimating 2x the size of the original text
|
|
totalTokens = [inputTotalTokens, outputTotalTokens]
|
|
return (t, totalTokens)
|
|
|
|
# Sub Vars
|
|
varResponse = subVars(t)
|
|
subbedT = varResponse[0]
|
|
|
|
# If there isn't any Japanese in the text just skip
|
|
if not re.search(r'[一-龠]+|[ぁ-ゔ]+|[ァ-ヴ]+|[\uFF00-\uFFEF]', subbedT):
|
|
return(t, [0,0])
|
|
|
|
# Characters
|
|
context = '```\
|
|
Game Characters:\
|
|
Character: ソル == Sol - Gender: Female\
|
|
Character: ェニ先生 == Eni-sensei - Gender: Female\
|
|
Character: 神泉 理央 == Kamiizumi Rio - Gender: Female\
|
|
Character: 吉祥寺 アリサ == Kisshouji Arisa - Gender: Female\
|
|
```'
|
|
|
|
# Prompt
|
|
if fullPromptFlag:
|
|
system = PROMPT
|
|
user = 'Line to Translate = ' + subbedT
|
|
else:
|
|
system = 'Output ONLY the english translation in the following format: `Translation: <ENGLISH_TRANSLATION>`'
|
|
user = 'Line to Translate = ' + subbedT
|
|
|
|
# Create Message List
|
|
msg = []
|
|
msg.append({"role": "system", "content": system})
|
|
msg.append({"role": "user", "content": context})
|
|
if isinstance(history, list):
|
|
for line in history:
|
|
msg.append({"role": "user", "content": line})
|
|
else:
|
|
msg.append({"role": "user", "content": history})
|
|
msg.append({"role": "user", "content": user})
|
|
|
|
response = openai.ChatCompletion.create(
|
|
temperature=0.1,
|
|
frequency_penalty=0.2,
|
|
presence_penalty=0.2,
|
|
model="gpt-3.5-turbo-1106",
|
|
messages=msg,
|
|
request_timeout=30,
|
|
)
|
|
|
|
# Save Translated Text
|
|
translatedText = response.choices[0].message.content
|
|
totalTokens = [response.usage.prompt_tokens, response.usage.completion_tokens]
|
|
|
|
# Resub Vars
|
|
translatedText = resubVars(translatedText, varResponse[1])
|
|
|
|
# Remove Placeholder Text
|
|
translatedText = translatedText.replace('English Translation: ', '')
|
|
translatedText = translatedText.replace('Translation: ', '')
|
|
translatedText = translatedText.replace('Line to Translate = ', '')
|
|
translatedText = translatedText.replace('Translation = ', '')
|
|
translatedText = translatedText.replace('Translate = ', '')
|
|
translatedText = translatedText.replace('English Translation:', '')
|
|
translatedText = translatedText.replace('Translation:', '')
|
|
translatedText = translatedText.replace('Line to Translate =', '')
|
|
translatedText = translatedText.replace('Translation =', '')
|
|
translatedText = translatedText.replace('Translate =', '')
|
|
translatedText = re.sub(r'Note:.*', '', translatedText)
|
|
translatedText = translatedText.replace('っ', '')
|
|
|
|
# Return Translation
|
|
if len(translatedText) > 15 * len(t) or "I'm sorry, but I'm unable to assist with that translation" in translatedText:
|
|
raise Exception
|
|
else:
|
|
return [translatedText, totalTokens] |