DazedTL/modules/alltext.py

from concurrent.futures import ThreadPoolExecutor, as_completed
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')

APICOST = .002 # Depends on the model https://openai.com/pricing
PROMPT = Path('prompt.txt').read_text(encoding='utf-8')
THREADS = 10 # For GPT4 rate limit will be hit if you have more than 1 thread.
LOCK = threading.Lock()
WIDTH = 60
LISTWIDTH = 60
MAXHISTORY = 10
ESTIMATE = ''
TOTALCOST = 0
TOKENS = 0
TOTALTOKENS = 0
NAMESLIST = []

#tqdm Globals
BAR_FORMAT='{l_bar}{bar:10}{r_bar}{bar:-10b}'
POSITION=0
LEAVE=False

def handleAllText(filename, estimate):
    global ESTIMATE, TOKENS, TOTALTOKENS, TOTALCOST
    ESTIMATE = estimate

    with open('translated/' + filename, 'w+t', newline='', encoding='utf-16-le') as writeFile:
        start = time.time()
        translatedData = openFiles(filename, writeFile)

    if estimate:
        # Print Result
        end = time.time()
        tqdm.write(getResultString(['', TOKENS, None], end - start, filename))
        TOTALCOST += TOKENS * .001 * APICOST
        TOTALTOKENS += TOKENS
        TOKENS = 0
        os.remove('translated/' + filename)
    
    else:
        # Print Result
        end = time.time()
        tqdm.write(getResultString(translatedData, end - start, filename))
        TOTALCOST += translatedData[1] * .001 * APICOST
        TOTALTOKENS += translatedData[1]

    return getResultString(['', TOTALTOKENS, None], end - start, 'TOTAL')

def openFiles(filename, writeFile):
    with open('files/' + filename, 'r', encoding='utf-16-le') as readFile, writeFile:
        translatedData = parseText(readFile, writeFile, filename)

    return translatedData

def getResultString(translatedData, translationTime, filename):
    # File Print String
    tokenString = Fore.YELLOW + '[' + str(translatedData[1]) + \
        ' Tokens/${:,.4f}'.format(translatedData[1] * .001 * APICOST) + ']'
    timeString = Fore.BLUE + '[' + str(round(translationTime, 1)) + 's]'

    if translatedData[2] == None:
        # Success
        return filename + ': ' + tokenString + timeString + Fore.GREEN + u' \u2713 ' + Fore.RESET

    else:
        # Fail
        try:
            raise translatedData[2]
        except Exception as e:
            errorString = str(e) + '|' + translatedData[3] + Fore.RED
            return filename + ': ' + tokenString + timeString + Fore.RED + u' \u2717 ' +\
                errorString + Fore.RESET
        
@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-4")
        tokens = len(enc.encode(t)) * 2 + len(enc.encode(str(history))) + len(enc.encode(PROMPT))
        return (t, tokens)
    
    # Sub Vars
    varResponse = subVars(t)
    subbedT = varResponse[0]

    # If there isn't any Japanese in the text just skip
    if not re.search(r'[<5B><>-<2D><>]+|[<5B><>-?]+|[<5B>@-<2D><>]+|[\uFF00-\uFFEF]', subbedT):
        return(t, 0)

    # Characters
    context = '```\
        Game Characters:\
        Character: <20>r<EFBFBD>m<EFBFBD><6D> <20><><EFBFBD>C == Ikenoue Takumi - Gender: Male\
        Character: <20><><EFBFBD>i <20><><EFBFBD>͂<EFBFBD> == Fukunaga Koharu - Gender: Female\
        Character: <20>_<EFBFBD><5F> <20><><EFBFBD><EFBFBD> == Kamiizumi Rio - Gender: Female\
        Character: <20>g<EFBFBD>ˎ<EFBFBD> <20>A<EFBFBD><41><EFBFBD>T == Kisshouji Arisa - Gender: Female\
        Character: <20>v<EFBFBD><76> <20>F<EFBFBD><46><EFBFBD>q == Kuga Yuriko - 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",
        messages=msg,
        request_timeout=30,
    )

    # Save Translated Text
    translatedText = response.choices[0].message.content
    tokens = response.usage.total_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('<EFBFBD><EFBFBD>', '')

    # 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, tokens]