762 lines
27 KiB
Python
762 lines
27 KiB
Python
# Libraries
|
||
import os
|
||
import re
|
||
import textwrap
|
||
import threading
|
||
import time
|
||
import traceback
|
||
import tiktoken
|
||
import openai
|
||
from pathlib import Path
|
||
from colorama import Fore
|
||
from dotenv import load_dotenv
|
||
from retry import retry
|
||
from tqdm import tqdm
|
||
|
||
# Open AI
|
||
load_dotenv()
|
||
if os.getenv("api").replace(" ", "") != "":
|
||
openai.base_url = os.getenv("api")
|
||
openai.organization = os.getenv("org")
|
||
openai.api_key = os.getenv("key")
|
||
|
||
# Globals
|
||
MODEL = os.getenv("model")
|
||
TIMEOUT = int(os.getenv("timeout"))
|
||
LANGUAGE = os.getenv("language").capitalize()
|
||
PROMPT = Path("prompt.txt").read_text(encoding="utf-8")
|
||
VOCAB = Path("vocab.txt").read_text(encoding="utf-8")
|
||
THREADS = int(os.getenv("threads"))
|
||
LOCK = threading.Lock()
|
||
WIDTH = int(os.getenv("width"))
|
||
LISTWIDTH = int(os.getenv("listWidth"))
|
||
NOTEWIDTH = 70
|
||
MAXHISTORY = 10
|
||
ESTIMATE = ""
|
||
TOKENS = [0, 0]
|
||
NAMESLIST = []
|
||
NAMES = False # Output a list of all the character names found
|
||
BRFLAG = False # If the game uses <br> instead
|
||
FIXTEXTWRAP = True # Overwrites textwrap
|
||
IGNORETLTEXT = False # Ignores all translated text.
|
||
MISMATCH = [] # Lists files that throw a mismatch error (Length of GPT list response is wrong)
|
||
|
||
# tqdm Globals
|
||
BAR_FORMAT = "{l_bar}{bar:10}{r_bar}{bar:-10b}"
|
||
POSITION = 0
|
||
LEAVE = False
|
||
|
||
# Regex - Need to change this if you want to translate from/to other languages. Default is Japanese Regex
|
||
LANGREGEX = r"[一-龠ぁ-ゔァ-ヴーａ-ｚＡ-Ｚ０-９\uFF61-\uFF9F]+"
|
||
|
||
# Pricing - Depends on the model https://openai.com/pricing
|
||
# Batch Size - GPT 3.5 Struggles past 15 lines per request. GPT4 struggles past 50 lines per request
|
||
# If you are getting a MISMATCH LENGTH error, lower the batch size.
|
||
if "gpt-3.5" in MODEL:
|
||
INPUTAPICOST = 3.00
|
||
OUTPUTAPICOST = 5.00
|
||
BATCHSIZE = 10
|
||
FREQUENCY_PENALTY = 0.2
|
||
elif "gpt-4" in MODEL:
|
||
INPUTAPICOST = 2.50
|
||
OUTPUTAPICOST = 10.00
|
||
BATCHSIZE = 30
|
||
FREQUENCY_PENALTY = 0.05
|
||
elif "deepseek" in MODEL:
|
||
INPUTAPICOST = 0.14
|
||
OUTPUTAPICOST = 1.10
|
||
BATCHSIZE = 30
|
||
FREQUENCY_PENALTY = 0.05
|
||
else:
|
||
INPUTAPICOST = float(os.getenv("input_cost"))
|
||
OUTPUTAPICOST = float(os.getenv("output_cost"))
|
||
BATCHSIZE = int(os.getenv("batchsize"))
|
||
FREQUENCY_PENALTY = float(os.getenv("frequency_penalty"))
|
||
|
||
|
||
def handleIris(filename, estimate):
|
||
global ESTIMATE
|
||
ESTIMATE = estimate
|
||
|
||
if ESTIMATE:
|
||
start = time.time()
|
||
translatedData = openFiles(filename)
|
||
|
||
# Print Result
|
||
end = time.time()
|
||
tqdm.write(getResultString(translatedData, end - start, filename))
|
||
with LOCK:
|
||
TOKENS[0] += translatedData[1][0]
|
||
TOKENS[1] += translatedData[1][1]
|
||
|
||
# Print Total
|
||
totalString = getResultString(["", TOKENS, None], end - start, "TOTAL")
|
||
|
||
# Print any errors on maps
|
||
if len(MISMATCH) > 0:
|
||
return totalString + Fore.RED + f"\nMismatch Errors: {MISMATCH}" + Fore.RESET
|
||
else:
|
||
return totalString
|
||
|
||
else:
|
||
try:
|
||
with open("translated/" + filename, "w", encoding="cp932", errors="ignore") as outFile:
|
||
start = time.time()
|
||
translatedData = openFiles(filename)
|
||
|
||
# Print Result
|
||
end = time.time()
|
||
outFile.writelines(translatedData[0])
|
||
tqdm.write(getResultString(translatedData, end - start, filename))
|
||
with LOCK:
|
||
TOKENS[0] += translatedData[1][0]
|
||
TOKENS[1] += translatedData[1][1]
|
||
except Exception:
|
||
traceback.print_exc()
|
||
return "Fail"
|
||
|
||
return getResultString(["", TOKENS, None], end - start, "TOTAL")
|
||
|
||
|
||
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] / 1000000) * INPUTAPICOST) + ((translatedData[1][1] / 1000000) * OUTPUTAPICOST))
|
||
+ "]"
|
||
)
|
||
timeString = Fore.BLUE + "[" + str(round(translationTime, 1)) + "s]"
|
||
|
||
if translatedData[2] == None:
|
||
# Success
|
||
return filename + ": " + totalTokenstring + timeString + Fore.GREEN + " \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 + " \u2717 " + errorString + Fore.RESET
|
||
|
||
|
||
def openFiles(filename):
|
||
with open("files/" + filename, "r", encoding="shift_jis") as readFile:
|
||
translatedData = parseIris(readFile, filename)
|
||
|
||
# Delete lines marked for deletion
|
||
finalData = []
|
||
for line in translatedData[0]:
|
||
if line != "\\d\n":
|
||
finalData.append(line)
|
||
translatedData[0] = finalData
|
||
|
||
return translatedData
|
||
|
||
|
||
def parseIris(readFile, filename):
|
||
totalTokens = [0, 0]
|
||
|
||
# Read File into data
|
||
data = readFile.readlines()
|
||
|
||
# Create Progress Bar
|
||
with tqdm(bar_format=BAR_FORMAT, position=POSITION, leave=LEAVE) as pbar:
|
||
pbar.desc = filename
|
||
|
||
try:
|
||
result = translateIris(data, pbar, filename, [])
|
||
totalTokens[0] += result[0]
|
||
totalTokens[1] += result[1]
|
||
except Exception as e:
|
||
traceback.print_exc()
|
||
return [data, totalTokens, e]
|
||
return [data, totalTokens, None]
|
||
|
||
|
||
def translateIris(data, pbar, filename, translatedList):
|
||
stringList = []
|
||
currentGroup = []
|
||
tokens = [0, 0]
|
||
speaker = ""
|
||
voice = False
|
||
global LOCK, ESTIMATE
|
||
i = 0
|
||
|
||
while i < len(data):
|
||
voice = False
|
||
speaker = ""
|
||
if "#MSGVOICE" in data[i]:
|
||
i += 1
|
||
voice = True
|
||
voiceVar = data[i]
|
||
if "#MSG," in data[i] or "#MSG\n" in data[i] or voice == True:
|
||
i += 1
|
||
# Speaker
|
||
if re.search(r'^ ?([^#\/."、。*!！（）\(\)\[\] \n]+)\n', data[i]) and len(data[i]) < 30:
|
||
match = re.search(r"(.*)", data[i])
|
||
if match != None:
|
||
speaker = match.group(1)
|
||
if speaker[0] == "\u3000":
|
||
speaker = speaker[1:]
|
||
response = getSpeaker(speaker, pbar, filename)
|
||
speaker = response[0]
|
||
tokens[0] += response[1][0]
|
||
tokens[1] += response[1][1]
|
||
if translatedList != []:
|
||
speaker = speaker.replace(" ", "\u3000")
|
||
data[i] = f"\u3000{speaker}\n"
|
||
else:
|
||
speaker = ""
|
||
i += 1
|
||
|
||
# Lines
|
||
match = re.search(r"(.*)", data[i])
|
||
if match != None and match.group(1) != "":
|
||
# Pass 1
|
||
if translatedList == []:
|
||
# Grab Consecutive Strings
|
||
jaString = data[i]
|
||
if data[i] != "\n":
|
||
if data[i][0] == "\u3000":
|
||
jaString = data[i][1:]
|
||
currentGroup.append(jaString)
|
||
i += 1
|
||
while data[i] != "\n":
|
||
jaString = data[i]
|
||
if data[i] != "\n":
|
||
jaString = data[i][1:]
|
||
currentGroup.append(jaString)
|
||
i += 1
|
||
|
||
# Join up 401 groups for better translation.
|
||
if len(currentGroup) > 0:
|
||
jaString = "".join(currentGroup)
|
||
currentGroup = []
|
||
|
||
# Remove any textwrap
|
||
jaString = jaString.replace("\n", " ")
|
||
|
||
# Temporarily convert spaces (For Textwrap Later)
|
||
jaString = jaString.replace("\u3000", " ")
|
||
|
||
# Add Speaker (If there is one)
|
||
if speaker != "":
|
||
jaString = f"{speaker}: {jaString}"
|
||
|
||
# Add String
|
||
stringList.append(jaString.strip())
|
||
|
||
# Pass 2
|
||
else:
|
||
# Insert Strings
|
||
while data[i] != "\n":
|
||
data.pop(i)
|
||
|
||
# Get Text
|
||
if translatedList:
|
||
translatedText = translatedList[0]
|
||
translatedList.pop(0)
|
||
if len(translatedList) <= 0:
|
||
translatedList = None
|
||
|
||
# Remove added speaker
|
||
translatedText = re.sub(r"^.+?:\s", "", translatedText)
|
||
|
||
# Textwrap
|
||
translatedText = textwrap.fill(translatedText, width=WIDTH)
|
||
translatedText = translatedText.replace("\n", "\n\u3000")
|
||
|
||
# Replace Whitespace and Commas
|
||
translatedText = translatedText.replace(", ", "、")
|
||
translatedText = translatedText.replace(",\u3000", "、")
|
||
translatedText = translatedText.replace(",", "、")
|
||
translatedText = translatedText.replace(" ", "\u3000")
|
||
|
||
# Set Data
|
||
# Game crashes on more than 3 lines. Will need to create a new MSG for long translations
|
||
if translatedText.count("\n") > 2:
|
||
# Split List
|
||
translatedTextList = splitNewlines(translatedText)
|
||
|
||
# MSG Voice
|
||
count = 0
|
||
for text in translatedTextList:
|
||
if count != 0:
|
||
if voice == True:
|
||
# MSG for each item in the list
|
||
data.insert(i, "#MSGVOICE,\n")
|
||
i += 1
|
||
data.insert(i, f"{voiceVar}")
|
||
i += 1
|
||
else:
|
||
data.insert(i, "#MSG,\n")
|
||
i += 1
|
||
if speaker:
|
||
data[i] = f"\u3000{speaker}\n"
|
||
i += 1
|
||
if text[0] == "\u3000":
|
||
data.insert(i, f"{text}\n")
|
||
else:
|
||
data.insert(i, f"\u3000{text}\n")
|
||
i += 1
|
||
count += 1
|
||
if data[i] != "\n":
|
||
data.insert(i, "\n")
|
||
data[i] = f"\n{data[i]}"
|
||
else:
|
||
data.insert(i, f"\u3000{translatedText}\n")
|
||
i += 1
|
||
if data[i] != "\n":
|
||
data[i] = f"\n{data[i]}"
|
||
|
||
elif "#SELECT" in data[i] and translatedList == []:
|
||
Iris = r"(.+?) +\d$"
|
||
i += 1
|
||
match = re.search(Iris, data[i])
|
||
if match:
|
||
choiceList = []
|
||
choiceList.append(match.group(1))
|
||
i += 1
|
||
match = re.search(Iris, data[i])
|
||
while match:
|
||
choiceList.append(match.group(1))
|
||
i += 1
|
||
match = re.search(Iris, data[i])
|
||
|
||
# Translate
|
||
question = stringList[len(stringList) - 1]
|
||
response = translateGPT(
|
||
choiceList,
|
||
f"Previous text for context: {question}\n\nThis will be a dialogue option",
|
||
True,
|
||
pbar,
|
||
filename,
|
||
)
|
||
tokens[0] += response[1][0]
|
||
tokens[1] += response[1][1]
|
||
choiceListTL = response[0]
|
||
|
||
# Set Data
|
||
i = i - len(choiceListTL)
|
||
for j in range(len(choiceListTL)):
|
||
# Replace Whitespace and Commas
|
||
choiceListTL[j] = choiceListTL[j].replace(", ", "、")
|
||
choiceListTL[j] = choiceListTL[j].replace(",\u3000", "、")
|
||
choiceListTL[j] = choiceListTL[j].replace(",", "、")
|
||
choiceListTL[j] = choiceListTL[j].replace(" ", "\u3000")
|
||
data[i] = data[i].replace(choiceList[j], choiceListTL[j])
|
||
i += 1
|
||
|
||
# Nothing relevant. Skip Line.
|
||
else:
|
||
i += 1
|
||
else:
|
||
i += 1
|
||
|
||
# EOF
|
||
if len(stringList) > 0:
|
||
# Set Progress
|
||
pbar.total = len(stringList)
|
||
pbar.refresh()
|
||
|
||
# Translate
|
||
response = translateGPT(stringList, "", True, pbar, filename)
|
||
tokens[0] += response[1][0]
|
||
tokens[1] += response[1][1]
|
||
translatedList = response[0]
|
||
|
||
# Set Strings
|
||
if len(stringList) == len(translatedList):
|
||
translateIris(data, pbar, filename, translatedList)
|
||
|
||
# Mismatch
|
||
else:
|
||
with LOCK:
|
||
if filename not in MISMATCH:
|
||
MISMATCH.append(filename)
|
||
return tokens
|
||
|
||
|
||
def splitNewlines(text):
|
||
parts = []
|
||
newline_count = 0 # Counts the number of newline characters encountered
|
||
start_index = 0 # Start index of the current string part
|
||
|
||
for i, char in enumerate(text):
|
||
if char == "\n":
|
||
newline_count += 1
|
||
if newline_count == 3:
|
||
# Append the string part from start_index to current index (inclusive)
|
||
parts.append(text[start_index : i + 1])
|
||
# Reset newline count and update start_index for the next string part
|
||
newline_count = 0
|
||
start_index = i + 1
|
||
|
||
# Edge case: if the text does not end with a newline, we still need to append the last part
|
||
if start_index < len(text):
|
||
parts.append(text[start_index:])
|
||
|
||
return parts
|
||
|
||
|
||
# Save some money and enter the character before translation
|
||
def getSpeaker(speaker, pbar, filename):
|
||
match speaker:
|
||
case "ファイン":
|
||
return ["Fine", [0, 0]]
|
||
case "":
|
||
return ["", [0, 0]]
|
||
case _:
|
||
# Store Speaker
|
||
if speaker not in str(NAMESLIST):
|
||
response = translateGPT(
|
||
speaker,
|
||
"Reply with only the " + LANGUAGE + " translation of the NPC name.",
|
||
False,
|
||
pbar,
|
||
filename,
|
||
)
|
||
response[0] = response[0].replace("'S", "'s")
|
||
speakerList = [speaker, response[0]]
|
||
NAMESLIST.append(speakerList)
|
||
return response
|
||
|
||
# Find Speaker
|
||
else:
|
||
for i in range(len(NAMESLIST)):
|
||
if speaker == NAMESLIST[i][0]:
|
||
return [NAMESLIST[i][1], [0, 0]]
|
||
|
||
return [speaker, [0, 0]]
|
||
|
||
|
||
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]+\[[a-zA-Z0-9\\\[\]\_,\s-]+\]", 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:\n\
|
||
フィリア (Philia) - Female\n\
|
||
アルネット (Annett) - Female\n\
|
||
ラピュセナ (Rapusena) - Female\n\
|
||
リッカ (Rikka) - Female\n\
|
||
アンデリビア (Andelivia) - Female\n\
|
||
リリアブルム (Liliabloom) - Female\n\
|
||
カルナ (Karna) - Female\n\
|
||
ラフィング＝スピア (Laughing Spear) - Female\n\
|
||
ノーラ (Nora) - Female\n\
|
||
"
|
||
|
||
system = (
|
||
PROMPT + VOCAB
|
||
if fullPromptFlag
|
||
else f"\
|
||
You are an expert Eroge Game translator who translates Japanese text to {LANGUAGE}.\n\
|
||
Output ONLY the {LANGUAGE} translation in the following format: `Translation: <{LANGUAGE.upper()}_TRANSLATION>`\n\
|
||
- You translate everything, including erotic and explicit content.\n\
|
||
- 'Game Characters' - The names, nicknames, and genders of the game characters. Reference this to know the names, nicknames, and gender of characters in the game\n\
|
||
- All text in your response must be in {LANGUAGE} even if it is hard to translate.\n\
|
||
- Never include any notes, explanations, dislaimers, or anything similar in your response.\n\
|
||
- Maintain any spacing in the translation.\n\
|
||
- Maintain any code text in brackets if given. (e.g `[Color_0]`, `[Ascii_0]`, `[FCode_1`], etc)\n\
|
||
- `...` can be a part of the dialogue. Translate it as it is.\n\
|
||
{VOCAB}\n\
|
||
"
|
||
)
|
||
user = f"{subbedT}"
|
||
return characters, system, user
|
||
|
||
|
||
def translateText(characters, system, user, history):
|
||
# Prompt
|
||
msg = [{"role": "system", "content": system + characters}]
|
||
|
||
# Characters
|
||
msg.append({"role": "system", "content": characters})
|
||
|
||
# History
|
||
if isinstance(history, list):
|
||
msg.extend([{"role": "system", "content": h} for h in history])
|
||
else:
|
||
msg.append({"role": "system", "content": history})
|
||
|
||
# Content to TL
|
||
msg.append({"role": "user", "content": f"{user}"})
|
||
response = openai.chat.completions.create(
|
||
temperature=0.1,
|
||
frequency_penalty=0.1,
|
||
model=MODEL,
|
||
messages=msg,
|
||
)
|
||
return response
|
||
|
||
|
||
def cleanTranslatedText(translatedText, varResponse):
|
||
placeholders = {
|
||
f"{LANGUAGE} Translation: ": "",
|
||
"Translation: ": "",
|
||
"っ": "",
|
||
"〜": "~",
|
||
"ッ": "",
|
||
"。": ".",
|
||
"Placeholder Text": "",
|
||
# Add more replacements as needed
|
||
}
|
||
for target, replacement in placeholders.items():
|
||
translatedText = translatedText.replace(target, replacement)
|
||
|
||
# Elongate Long Dashes (Since GPT Ignores them...)
|
||
translatedText = elongateCharacters(translatedText)
|
||
translatedText = resubVars(translatedText, varResponse[1])
|
||
return translatedText
|
||
|
||
|
||
def elongateCharacters(text):
|
||
# Define a pattern to match one character followed by one or more `ー` characters
|
||
# Using a positive lookbehind assertion to capture the preceding character
|
||
pattern = r"(?<=(.))ー+"
|
||
|
||
# Define a replacement function that elongates the captured character
|
||
def repl(match):
|
||
char = match.group(1) # The character before the ー sequence
|
||
count = len(match.group(0)) - 1 # Number of ー characters
|
||
return char * count # Replace ー sequence with the character repeated
|
||
|
||
# Use re.sub() to replace the pattern in the text
|
||
return re.sub(pattern, repl, text)
|
||
|
||
|
||
def extractTranslation(translatedTextList, is_list):
|
||
pattern = r"`?<[Ll]ine\d+>([\\]*.*?[\\]*?)<\/?[Ll]ine\d+>`?"
|
||
# If it's a batch (i.e., list), extract with tags; otherwise, return the single item.
|
||
if is_list:
|
||
matchList = re.findall(pattern, translatedTextList)
|
||
return matchList
|
||
else:
|
||
matchList = re.findall(pattern, translatedTextList)
|
||
return matchList[0][0] if matchList else translatedTextList
|
||
|
||
|
||
def countTokens(characters, system, user, history):
|
||
inputTotalTokens = 0
|
||
outputTotalTokens = 0
|
||
enc = tiktoken.encoding_for_model("gpt-4")
|
||
|
||
# Input
|
||
if isinstance(history, list):
|
||
for line in history:
|
||
inputTotalTokens += len(enc.encode(line))
|
||
else:
|
||
inputTotalTokens += len(enc.encode(history))
|
||
inputTotalTokens += len(enc.encode(system))
|
||
inputTotalTokens += len(enc.encode(characters))
|
||
inputTotalTokens += len(enc.encode(user))
|
||
|
||
# Output
|
||
outputTotalTokens += round(len(enc.encode(user)) * 2)
|
||
|
||
return [inputTotalTokens, outputTotalTokens]
|
||
|
||
|
||
@retry(exceptions=Exception, tries=5, delay=5)
|
||
def translateGPT(text, history, fullPromptFlag, pbar, filename):
|
||
mismatch = False
|
||
totalTokens = [0, 0]
|
||
if isinstance(text, list):
|
||
tList = batchList(text, BATCHSIZE)
|
||
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"`<Line{i}>{item}</Line{i}>`" for i, item in enumerate(tItem)])
|
||
payload = re.sub(r"(<Line\d+)(><)(\/Line\d+>)", r"\1>Placeholder Text<\3", payload)
|
||
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
|
||
|
||
# Create Message
|
||
characters, system, user = createContext(fullPromptFlag, subbedT)
|
||
|
||
# Calculate Estimate
|
||
if ESTIMATE:
|
||
estimate = countTokens(characters, system, user, history)
|
||
totalTokens[0] += estimate[0]
|
||
totalTokens[1] += estimate[1]
|
||
continue
|
||
|
||
# Translating
|
||
response = translateText(characters, system, user, history)
|
||
translatedText = response.choices[0].message.content
|
||
totalTokens[0] += response.usage.prompt_tokens
|
||
totalTokens[1] += response.usage.completion_tokens
|
||
|
||
# Formatting
|
||
translatedText = cleanTranslatedText(translatedText, varResponse)
|
||
if isinstance(tItem, list):
|
||
extractedTranslations = extractTranslation(translatedText, True)
|
||
if len(tItem) != len(extractedTranslations):
|
||
# Mismatch. Try Again
|
||
response = translateText(characters, system, user, history)
|
||
translatedText = response.choices[0].message.content
|
||
totalTokens[0] += response.usage.prompt_tokens
|
||
totalTokens[1] += response.usage.completion_tokens
|
||
|
||
# Formatting
|
||
translatedText = cleanTranslatedText(translatedText, varResponse)
|
||
if isinstance(tItem, list):
|
||
extractedTranslations = extractTranslation(translatedText, True)
|
||
if len(tItem) == len(extractedTranslations):
|
||
tList[index] = extractedTranslations
|
||
else:
|
||
MISMATCH.append(filename)
|
||
else:
|
||
tList[index] = extractedTranslations
|
||
|
||
# Create History
|
||
history = tList[index] # Update history if we have a list
|
||
pbar.update(len(tList[index]))
|
||
|
||
else:
|
||
# Ensure we're passing a single string to extractTranslation
|
||
extractedTranslations = extractTranslation(translatedText, False)
|
||
tList[index] = extractedTranslations
|
||
|
||
# Combine if multilist
|
||
if isinstance(tList[0], list):
|
||
tList = [t for sublist in tList for t in sublist]
|
||
|
||
# Return
|
||
if format == "json":
|
||
return [tList, totalTokens]
|
||
else:
|
||
return [tList[0], totalTokens]
|