581 lines
20 KiB
Python
581 lines
20 KiB
Python
# Libraries
|
||
import json
|
||
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 handleAnim(filename, estimate):
|
||
global ESTIMATE
|
||
totalTokens = [0, 0]
|
||
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]
|
||
|
||
# Print Total
|
||
totalString = getResultString(["", totalTokens, 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="utf-8", newline="\n") as outFile:
|
||
start = time.time()
|
||
translatedData = openFiles(filename)
|
||
|
||
# Print Result
|
||
end = time.time()
|
||
json.dump(translatedData[0], outFile, ensure_ascii=False, indent=4)
|
||
tqdm.write(getResultString(translatedData, end - start, filename))
|
||
with LOCK:
|
||
totalTokens[0] += translatedData[1][0]
|
||
totalTokens[1] += translatedData[1][1]
|
||
except Exception:
|
||
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] / 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 parseJSON(data, filename):
|
||
keys = list(data.keys())
|
||
batches = [keys[i : i + BATCHSIZE] for i in range(0, len(keys), BATCHSIZE)]
|
||
totalTokens = [0, 0]
|
||
totalLines = 0
|
||
totalLines = len(batches)
|
||
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(batches, data, pbar)
|
||
totalTokens[0] += result[0]
|
||
totalTokens[1] += result[1]
|
||
except Exception as e:
|
||
traceback.print_exc()
|
||
return [data, totalTokens, e]
|
||
return [data, totalTokens, None]
|
||
|
||
|
||
def translateJSON(keys, data, pbar):
|
||
translatedBatch = []
|
||
textHistory = []
|
||
tokens = [0, 0]
|
||
|
||
for batch in keys:
|
||
# Save Batch
|
||
originalBatch = batch.copy()
|
||
|
||
# If there isn't any Japanese in the text just skip
|
||
needTL = False
|
||
for i in range(len(batch)):
|
||
t = data[batch[i]]
|
||
if re.search(r"[一-龠ぁ-ゔァ-ヴーａ-ｚＡ-Ｚ０-９]+", t) or t == "":
|
||
needTL = True
|
||
if needTL is False and IGNORETLTEXT is True:
|
||
pbar.update(1)
|
||
continue
|
||
|
||
# Remove any textwrap and Furigana
|
||
for i in range(len(batch)):
|
||
if FIXTEXTWRAP == True:
|
||
# Textwrap
|
||
data[originalBatch[i]] = data[originalBatch[i]].replace("@b", " ")
|
||
|
||
# Furigana
|
||
rcodeMatch = re.findall(r"(@\[(.+?):.+?\])", batch[i])
|
||
if len(rcodeMatch) > 0:
|
||
for match in rcodeMatch:
|
||
batch[i] = batch[i].replace(match[0], match[1])
|
||
|
||
# Translate
|
||
if needTL is True:
|
||
response = translateGPT(batch, textHistory, True)
|
||
tokens[0] += response[1][0]
|
||
tokens[1] += response[1][1]
|
||
translatedBatch = response[0]
|
||
else:
|
||
for i in range(len(originalBatch)):
|
||
translatedBatch.append(data[originalBatch[i]])
|
||
|
||
# Format and Set Text
|
||
if len(batch) == len(translatedBatch):
|
||
for i in range(len(translatedBatch)):
|
||
# Remove added speaker
|
||
translatedText = translatedBatch[i]
|
||
translatedText = re.sub(r"^.+?\s\|\s?", "", translatedText)
|
||
|
||
# Textwrap
|
||
if "@n" in translatedText:
|
||
match = re.search(r".*@n(.*)", translatedText)
|
||
if match != None:
|
||
tlText = match.group(1)
|
||
tlText = textwrap.fill(tlText, width=WIDTH)
|
||
tlText = tlText.replace("\n", "@b")
|
||
translatedText = translatedText.replace(match.group(1), tlText)
|
||
|
||
elif "@b" not in translatedText:
|
||
translatedText = textwrap.fill(translatedText, width=WIDTH)
|
||
translatedText = translatedText.replace("\n", "@b")
|
||
|
||
# Set Data
|
||
data[originalBatch[i]] = translatedText
|
||
textHistory = translatedBatch
|
||
translatedBatch.clear()
|
||
# Mismatch, Skip Batch
|
||
else:
|
||
MISMATCH.append(batch)
|
||
pbar.update(1)
|
||
continue
|
||
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]+\[[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\
|
||
達也 (Tatsuya) - Male\n\
|
||
香織 (Kaori) - Female\n\
|
||
岩瀬 (Iwase)\n\
|
||
万蔵 (Manzou) - Male\n\
|
||
結奈 (Yuuna) - Female\n\
|
||
茅部 (Kayabe)\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, penalty):
|
||
# 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,
|
||
frequency_penalty=penalty,
|
||
model=MODEL,
|
||
messages=msg,
|
||
)
|
||
return response
|
||
|
||
|
||
def cleanTranslatedText(translatedText, varResponse):
|
||
placeholders = {
|
||
f"{LANGUAGE} Translation: ": "",
|
||
"Translation: ": "",
|
||
"っ": "",
|
||
"〜": "~",
|
||
"ッ": "",
|
||
"。": ".",
|
||
"Placeholder Text": "",
|
||
"é": "e",
|
||
"—": "-",
|
||
"ū": "u",
|
||
# 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)) * 3)
|
||
|
||
return [inputTotalTokens, outputTotalTokens]
|
||
|
||
|
||
@retry(exceptions=Exception, tries=5, delay=5)
|
||
def translateGPT(text, history, fullPromptFlag):
|
||
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, 0.02)
|
||
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)
|
||
tList[index] = extractedTranslations
|
||
if len(tItem) != len(extractedTranslations):
|
||
# Mismatch. Try Again
|
||
response = translateText(characters, system, user, history, 0.1)
|
||
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)
|
||
tList[index] = extractedTranslations
|
||
if len(tItem) != len(extractedTranslations):
|
||
mismatch = True # Just here for breakpoint
|
||
|
||
# Create History
|
||
if not mismatch:
|
||
history = extractedTranslations[-10:] # Update history if we have a list
|
||
else:
|
||
history = text[-10:]
|
||
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]
|