Changes to txt script
This commit is contained in:
parent
446dd1d83f
commit
e1e13bbe4d
2 changed files with 118 additions and 63 deletions
|
|
@ -25,7 +25,7 @@ APICOST = .002 # Depends on the model https://openai.com/pricing
|
|||
PROMPT = Path('prompt.txt').read_text(encoding='utf-8')
|
||||
THREADS = 20 # For GPT4 rate limit will be hit if you have more than 1 thread.
|
||||
LOCK = threading.Lock()
|
||||
WIDTH = 60
|
||||
WIDTH = 50
|
||||
LISTWIDTH = 60
|
||||
MAXHISTORY = 10
|
||||
ESTIMATE = ''
|
||||
|
|
@ -44,7 +44,7 @@ CODE401 = True
|
|||
CODE405 = False
|
||||
CODE102 = True
|
||||
CODE122 = False
|
||||
CODE101 = False
|
||||
CODE101 = True
|
||||
CODE355655 = False
|
||||
CODE357 = False
|
||||
CODE657 = False
|
||||
|
|
@ -216,22 +216,19 @@ def translateNote(event, regex):
|
|||
|
||||
match = re.findall(regex, jaString, re.DOTALL)
|
||||
if match:
|
||||
jaString = match[0]
|
||||
# Need to remove outside code
|
||||
jaString = re.sub(r'^[^一-龠ぁ-ゔァ-ヴー\<\>【】]+', '', jaString)
|
||||
jaString = re.sub(r'[^一-龠ぁ-ゔァ-ヴー\<\>【】。！？]+$', '', jaString)
|
||||
oldjaString = jaString
|
||||
|
||||
oldJAString = match[0]
|
||||
# Remove any textwrap
|
||||
jaString = re.sub(r'\n', ' ', jaString)
|
||||
|
||||
response = translateGPT(jaString, '', True)
|
||||
jaString = re.sub(r'\n', ' ', oldJAString)
|
||||
|
||||
# Translate
|
||||
response = translateGPT(jaString, 'Reply with the English translation of the NPC name.', True)
|
||||
translatedText = response[0]
|
||||
|
||||
# Textwrap
|
||||
translatedText = textwrap.fill(translatedText, width=LISTWIDTH)
|
||||
|
||||
event['note'] = event['note'].replace(oldjaString, translatedText)
|
||||
translatedText = translatedText.replace('\"', '')
|
||||
event['note'] = event['note'].replace(oldJAString, translatedText)
|
||||
return response[1]
|
||||
return 0
|
||||
|
||||
|
|
@ -434,10 +431,12 @@ def searchNames(name, pbar, context):
|
|||
responseList.append(translateGPT(name['description'], '', True))
|
||||
else:
|
||||
responseList.append(['', 0])
|
||||
if 'hint' in name['note']:
|
||||
tokens += translateNote(name, r'<hint:([^>]*)>')
|
||||
|
||||
if 'Enemies' in context:
|
||||
if 'desc1' in name['note']:
|
||||
tokens += translateNote(name, r'<desc1:([^>]*)>')
|
||||
if 'taneoya' in name['note']:
|
||||
tokens += translateNote(name, r'<taneoya:([^>]*)>')
|
||||
|
||||
if 'desc2' in name['note']:
|
||||
tokens += translateNote(name, r'<desc2:([^>]*)>')
|
||||
|
|
@ -506,8 +505,6 @@ def searchCodes(page, pbar):
|
|||
# if jaString.startswith(' '):
|
||||
# jaString = jaString + ': '
|
||||
|
||||
jaString = re.sub(r'([\u3000-\uffef])\1{3,}', r'\1\1\1', jaString)
|
||||
|
||||
# Using this to keep track of 401's in a row. Throws IndexError at EndOfList (Expected Behavior)
|
||||
currentGroup.append(jaString)
|
||||
|
||||
|
|
@ -517,7 +514,6 @@ def searchCodes(page, pbar):
|
|||
i += 1
|
||||
|
||||
jaString = codeList[i]['parameters'][0]
|
||||
jaString = re.sub(r'([\u3000-\uffef])\1{3,}', r'\1\1\1', jaString)
|
||||
currentGroup.append(jaString)
|
||||
|
||||
# Join up 401 groups for better translation.
|
||||
|
|
@ -532,7 +528,6 @@ def searchCodes(page, pbar):
|
|||
response = translateGPT(match[0][1], 'Reply with only the english translation of the NPC name', True)
|
||||
tokens += response[1]
|
||||
speaker = response[0].strip('.')
|
||||
speakerVar = match[0][0].replace(match[0][1], response[0])
|
||||
finalJAString = finalJAString.replace(match[0][0], '')
|
||||
|
||||
# Put names in list
|
||||
|
|
@ -571,7 +566,7 @@ def searchCodes(page, pbar):
|
|||
textHistory.append('\"' + translatedText + '\"')
|
||||
|
||||
# Remove added speaker
|
||||
translatedText = translatedText.replace(speaker + ': ', speakerVar)
|
||||
translatedText = re.sub(r'^.+?:\s', '', translatedText)
|
||||
speaker = ''
|
||||
speakerVar = ''
|
||||
else:
|
||||
|
|
@ -763,6 +758,9 @@ def searchCodes(page, pbar):
|
|||
# Set Data
|
||||
speaker = translatedText
|
||||
codeList[i]['parameters'][4] = translatedText
|
||||
if NAMES == True and speaker not in NAMESLIST:
|
||||
with LOCK:
|
||||
NAMESLIST.append(speaker)
|
||||
|
||||
## Event Code: 355 or 655 Scripts [Optional]
|
||||
if (codeList[i]['code'] == 355) and CODE355655 == True:
|
||||
|
|
@ -1089,16 +1087,16 @@ def searchSS(state, pbar):
|
|||
tokens = 0
|
||||
|
||||
# Name
|
||||
nameResponse = translateGPT(state['name'], 'Reply with only the english translation of the RPG Skill name.', False) if 'name' in state else ''
|
||||
nameResponse = translateGPT(state['name'], 'Reply with only the english translation of the RPG Skill name.', True) if 'name' in state else ''
|
||||
|
||||
# Description
|
||||
descriptionResponse = translateGPT(state['description'], 'Reply with only the english translation of the description.', False) if 'description' in state else ''
|
||||
descriptionResponse = translateGPT(state['description'], 'Reply with only the english translation of the description.', True) if 'description' in state else ''
|
||||
|
||||
# Messages
|
||||
message1Response = translateGPT('Taro' + state['message1'], 'reply with only the gender neutral english translation of the text.', False) if 'message1' in state else ''
|
||||
message2Response = translateGPT(state['message2'], 'reply with only the english translation of the text.', False) if 'message2' in state else ''
|
||||
message3Response = translateGPT(state['message3'], 'reply with only the english translation of the text.', False) if 'message3' in state else ''
|
||||
message4Response = translateGPT(state['message4'], 'reply with only the english translation of the text.', False) if 'message4' in state else ''
|
||||
message1Response = translateGPT('Taro' + state['message1'], 'reply with only the gender neutral english translation of the action.', True) if 'message1' in state else ''
|
||||
message2Response = translateGPT(state['message2'], 'reply with only the english translation of the text.', True) if 'message2' in state else ''
|
||||
message3Response = translateGPT(state['message3'], 'reply with only the english translation of the text.', True) if 'message3' in state else ''
|
||||
message4Response = translateGPT(state['message4'], 'reply with only the english translation of the text.', True) if 'message4' in state else ''
|
||||
|
||||
# if 'note' in state:
|
||||
if 'DOBBY' in state['note']:
|
||||
|
|
@ -1243,15 +1241,16 @@ def translateGPT(t, history, fullPromptFlag):
|
|||
return(t, 0)
|
||||
|
||||
"""Translate text using GPT"""
|
||||
context = 'Eroge Names Context: ミカエル == Mikael | Female, ミカ == Mika | Female, ベルゼビュート == Beelzebuth | Female, ベル == Bel | Female, アズラエル == Azriel | Female, アズ == Az | Female, フレイア == Freya | Female'
|
||||
context = 'Eroge Names Context: カレン == Karen | Female, エリス == Eris | Female, コレット == Colette | Female, テオ == Teo | Male, メイヴィス == Mavis | Female, '
|
||||
if fullPromptFlag:
|
||||
system = PROMPT
|
||||
user = 'Current Text to Translate: ' + subbedT
|
||||
user = 'Line to Translate: ' + subbedT
|
||||
else:
|
||||
system = 'You are an expert translator who translates everything to English. Reply with only the English Translation of the text.'
|
||||
user = 'Current Text to Translate: ' + subbedT
|
||||
user = 'Line to Translate: ' + subbedT
|
||||
response = openai.ChatCompletion.create(
|
||||
temperature=0,
|
||||
frequency_penalty=1,
|
||||
model="gpt-3.5-turbo",
|
||||
messages=[
|
||||
{"role": "system", "content": system},
|
||||
|
|
@ -1272,10 +1271,10 @@ def translateGPT(t, history, fullPromptFlag):
|
|||
# Remove Placeholder Text
|
||||
translatedText = translatedText.replace('English Translation: ', '')
|
||||
translatedText = translatedText.replace('Translation: ', '')
|
||||
translatedText = translatedText.replace('Current Text to Translate: ', '')
|
||||
translatedText = translatedText.replace('Line to Translate: ', '')
|
||||
translatedText = translatedText.replace('English Translation:', '')
|
||||
translatedText = translatedText.replace('Translation:', '')
|
||||
translatedText = translatedText.replace('Current Text to Translate:', '')
|
||||
translatedText = translatedText.replace('Line to Translate:', '')
|
||||
|
||||
# Return Translation
|
||||
if len(translatedText) > 15 * len(t) or "I'm sorry, but I'm unable to assist with that translation" in translatedText:
|
||||
|
|
|
|||
|
|
@ -131,40 +131,47 @@ def translateText(data, pbar):
|
|||
tokens = 0
|
||||
speaker = ''
|
||||
speakerFlag = False
|
||||
currentGroup = []
|
||||
syncIndex = 0
|
||||
|
||||
for i in range(len(data)):
|
||||
if '◆' in data[i]:
|
||||
jaString = data[i]
|
||||
if i != syncIndex:
|
||||
continue
|
||||
|
||||
match = re.findall(r'm\[[0-9]+\] = \"(.+?)\"', data[i])
|
||||
if len(match) > 0:
|
||||
jaString = match[0]
|
||||
|
||||
### Translate
|
||||
# Remove any textwrap
|
||||
jaString = re.sub(r'\\n', ' ', jaString)
|
||||
|
||||
# Check if speaker
|
||||
if '◆A' in jaString:
|
||||
speakerFlag = True
|
||||
# Grab Speaker
|
||||
speakerMatch = re.findall(r's\[[0-9]+\] = \"(.+?)[／\"]', data[i-1])
|
||||
if len(match) != 0:
|
||||
response = translateGPT(speakerMatch[0], 'Reply with only the english translation of the NPC name', True)
|
||||
tokens += response[1]
|
||||
speaker = response[0].strip('.')
|
||||
|
||||
# Need to remove outside code and put it back later
|
||||
startString = re.search(r'^◆[a-zA-Z0-9]+◆', jaString)
|
||||
jaString = re.sub(r'^◆[a-zA-Z0-9]+◆', '', jaString)
|
||||
endString = re.search(r'\n$', jaString)
|
||||
jaString = re.sub(r'\n$', '', jaString)
|
||||
if startString is None: startString = ''
|
||||
else: startString = startString.group()
|
||||
if endString is None: endString = ''
|
||||
else: endString = endString.group()
|
||||
|
||||
# Remove Repeating Chars
|
||||
jaString = re.sub(r'([\u3000-\uffef])\1{1,}', r'\1', jaString)
|
||||
# Grab rest of the messages
|
||||
currentGroup.append(jaString)
|
||||
start = i
|
||||
while (re.search(r'm\[[0-9]+\] = \"(.+?)\"', data[i+1]) != None):
|
||||
i+=1
|
||||
match = re.findall(r'm\[[0-9]+\] = \"(.+?)\"', data[i])
|
||||
currentGroup.append(match[0])
|
||||
finalJAString = ''.join(currentGroup)
|
||||
|
||||
# Translate
|
||||
if speaker != '':
|
||||
response = translateGPT(jaString, 'Previous Text for Context: ' + ' '.join(textHistory) \
|
||||
+ '\n\n\n###\n\n\nCurrent Speaker: ' + speaker, True)
|
||||
response = translateGPT(f'{speaker}: {finalJAString}', 'Previous Text for Context: ' + ' '.join(textHistory), True)
|
||||
else:
|
||||
response = translateGPT(jaString, 'Previous Text for Context: ' + ' '.join(textHistory), True)
|
||||
response = translateGPT(finalJAString, 'Previous Text for Context: ' + ' '.join(textHistory), True)
|
||||
tokens += response[1]
|
||||
translatedText = response[0]
|
||||
|
||||
# Remove added speaker
|
||||
translatedText = re.sub(r'^.+?:\s', '', translatedText)
|
||||
|
||||
# TextHistory is what we use to give GPT Context, so thats appended here.
|
||||
# rawTranslatedText = re.sub(r'[\\<>]+[a-zA-Z]+\[[a-zA-Z0-9]+\]', '', translatedText)
|
||||
|
|
@ -176,6 +183,7 @@ def translateText(data, pbar):
|
|||
# Keep textHistory list at length maxHistory
|
||||
if len(textHistory) > maxHistory:
|
||||
textHistory.pop(0)
|
||||
currentGroup = []
|
||||
|
||||
# Textwrap
|
||||
translatedText = textwrap.fill(translatedText, width=WIDTH)
|
||||
|
|
@ -193,45 +201,93 @@ def translateText(data, pbar):
|
|||
speakerFlag = False
|
||||
|
||||
# Write
|
||||
data[i] = startString + translatedText + endString
|
||||
data[i] = translatedText
|
||||
syncIndex = i + 1
|
||||
pbar.update()
|
||||
return [data, tokens]
|
||||
|
||||
def subVars(jaString):
|
||||
varRegex = r'\\+[a-zA-Z]+\[[0-9a-zA-Z\\\[\]]+\]+|[\\]+[#|]+|\\+[\\\[\]\.<>a-zA-Z0-9]+'
|
||||
count = 0
|
||||
|
||||
varList = re.findall(varRegex, jaString)
|
||||
if len(varList) != 0:
|
||||
for var in varList:
|
||||
jaString = jaString.replace(var, '<x' + str(count) + '>')
|
||||
count += 1
|
||||
|
||||
return [jaString, varList]
|
||||
|
||||
def resubVars(translatedText, varList):
|
||||
count = 0
|
||||
|
||||
if len(varList) != 0:
|
||||
for var in varList:
|
||||
translatedText = translatedText.replace('<x' + 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):
|
||||
with LOCK:
|
||||
# If ESTIMATE is True just count this as an execution and return.
|
||||
if ESTIMATE:
|
||||
global TOKENS
|
||||
enc = tiktoken.encoding_for_model("gpt-3.5-turbo")
|
||||
enc = tiktoken.encoding_for_model("gpt-3.5-turbo-0613")
|
||||
TOKENS += len(enc.encode(t)) * 2 + len(enc.encode(history)) + len(enc.encode(PROMPT))
|
||||
return (t, 0)
|
||||
|
||||
# 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]+', t):
|
||||
if not re.search(r'[一-龠]+|[ぁ-ゔ]+|[ァ-ヴー]+', subbedT):
|
||||
return(t, 0)
|
||||
|
||||
"""Translate text using GPT"""
|
||||
context = 'Eroge Names Context: カレン == Karen | Female, エリス == Eris | Female, コレット == Colette | Female, テオ == Teo | Male, メイヴィス == Mavis | Female, '
|
||||
if fullPromptFlag:
|
||||
system = PROMPT + history
|
||||
system = PROMPT
|
||||
user = 'Line to Translate: ' + subbedT
|
||||
else:
|
||||
system = 'You are going to pretend to be Japanese visual novel translator, \
|
||||
editor, and localizer. ' + history
|
||||
system = 'You are an expert translator who translates everything to English. Reply with only the English Translation of the text.'
|
||||
user = 'Line to Translate: ' + subbedT
|
||||
response = openai.ChatCompletion.create(
|
||||
temperature=0,
|
||||
frequency_penalty=1,
|
||||
model="gpt-3.5-turbo",
|
||||
messages=[
|
||||
{"role": "system", "content": system},
|
||||
{"role": "user", "content": "Text to Translate: " + t}
|
||||
{"role": "user", "content": context},
|
||||
{"role": "user", "content": history},
|
||||
{"role": "user", "content": user}
|
||||
],
|
||||
request_timeout=30,
|
||||
)
|
||||
|
||||
# Make sure translation didn't wonk out
|
||||
mlen=len(response.choices[0].message.content)
|
||||
elnt=10*len(t)
|
||||
if len(response.choices[0].message.content) > 9 * len(t):
|
||||
# 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('English Translation:', '')
|
||||
translatedText = translatedText.replace('Translation:', '')
|
||||
translatedText = translatedText.replace('Line to Translate:', '')
|
||||
|
||||
# Return Translation
|
||||
if len(translatedText) > 15 * len(t) or "I'm sorry, but I'm unable to assist with that translation" in translatedText:
|
||||
return [t, response.usage.total_tokens]
|
||||
else:
|
||||
return [response.choices[0].message.content, response.usage.total_tokens]
|
||||
|
||||
return [translatedText, tokens]
|
||||
Loading…
Reference in a new issue