From 3b7215fe77c5aa00399260d676fd18bc7390bb67 Mon Sep 17 00:00:00 2001 From: Dazed Date: Wed, 6 Dec 2023 17:46:35 -0600 Subject: [PATCH] Changes to TranslateGPT --- modules/anim.py | 68 +++++++++++++++++++++++++++-------------- modules/rpgmakermvmz.py | 7 +++-- prompt.example | 31 ++++++++++++------- 3 files changed, 69 insertions(+), 37 deletions(-) diff --git a/modules/anim.py b/modules/anim.py index ef8480e..26554b4 100644 --- a/modules/anim.py +++ b/modules/anim.py @@ -30,7 +30,7 @@ NAMESLIST = [] NAMES = False # Output a list of all the character names found BRFLAG = False # If the game uses
instead FIXTEXTWRAP = True # Overwrites textwrap -IGNORETLTEXT = True # Ignores all translated text. +IGNORETLTEXT = False # Ignores all translated text. MISMATCH = [] # Lists files that throw a mismatch error (Length of GPT list response is wrong) #tqdm Globals @@ -66,7 +66,14 @@ def handleAnim(filename, estimate): totalTokens[0] += translatedData[1][0] totalTokens[1] += translatedData[1][1] - return getResultString(['', totalTokens, None], end - start, 'TOTAL') + # 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: @@ -139,38 +146,50 @@ def parseJSON(data, 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 translateJSON(keys, data, pbar): + translatedBatch = [] textHistory = [] tokens = [0, 0] for batch in keys: # Save Batch - originalBatch = batch + originalBatch = batch.copy() # If there isn't any Japanese in the text just skip - if IGNORETLTEXT is True: - needTL = False - for i in range(len(batch)): - t = data[batch[i]] - if t == "": - needTL = True - if needTL is False: - pbar.update(1) - continue + 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 - if FIXTEXTWRAP == True: - for i in range(len(batch)): - data[originalBatch[i]] = data[originalBatch[i]].replace('\n', ' ') + # 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 - response = translateGPT(batch, textHistory, True) - tokens[0] += response[1][0] - tokens[1] += response[1][1] - translatedBatch = response[0] + 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): @@ -181,12 +200,14 @@ def translateJSON(keys, data, pbar): translatedText = re.sub(r'^.+?\s\|\s?', '', translatedText) # Textwrap - if '\n' not in translatedText: + if '@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) @@ -356,6 +377,7 @@ def cleanTranslatedText(translatedText, varResponse): placeholders = { f'{LANGUAGE} Translation: ': '', 'Translation: ': '', + 'っ': '', # Add more replacements as needed } for target, replacement in placeholders.items(): @@ -365,7 +387,7 @@ def cleanTranslatedText(translatedText, varResponse): return [line for line in translatedText.split('\n') if line] def extractTranslation(translatedTextList, is_list): - pattern = r'L(\d+) - (.*)' + pattern = r'(.*)' # If it's a batch (i.e., list), extract with tags; otherwise, return the single item. if is_list: return [re.findall(pattern, line)[0][1] for line in translatedTextList if re.search(pattern, line)] @@ -409,7 +431,7 @@ def translateGPT(text, history, fullPromptFlag): 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'L{i} - {item}' for i, item in enumerate(tItem)]) + payload = '\n'.join([f'{item}' for i, item in enumerate(tItem)]) varResponse = subVars(payload) subbedT = varResponse[0] else: diff --git a/modules/rpgmakermvmz.py b/modules/rpgmakermvmz.py index 51998d7..3fe1e9e 100644 --- a/modules/rpgmakermvmz.py +++ b/modules/rpgmakermvmz.py @@ -44,7 +44,7 @@ if 'gpt-3.5' in MODEL: elif 'gpt-4' in MODEL: INPUTAPICOST = .01 OUTPUTAPICOST = .03 - BATCHSIZE = 50 + BATCHSIZE = 40 #tqdm Globals BAR_FORMAT='{l_bar}{bar:10}{r_bar}{bar:-10b}' @@ -1880,6 +1880,7 @@ def cleanTranslatedText(translatedText, varResponse): placeholders = { f'{LANGUAGE} Translation: ': '', 'Translation: ': '', + 'っ': '', # Add more replacements as needed } for target, replacement in placeholders.items(): @@ -1889,7 +1890,7 @@ def cleanTranslatedText(translatedText, varResponse): return [line for line in translatedText.split('\n') if line] def extractTranslation(translatedTextList, is_list): - pattern = r'L(\d+) - (.*)' + pattern = r'(.*)' # If it's a batch (i.e., list), extract with tags; otherwise, return the single item. if is_list: return [re.findall(pattern, line)[0][1] for line in translatedTextList if re.search(pattern, line)] @@ -1933,7 +1934,7 @@ def translateGPT(text, history, fullPromptFlag): 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'L{i} - {item}' for i, item in enumerate(tItem)]) + payload = '\n'.join([f'{item}' for i, item in enumerate(tItem)]) varResponse = subVars(payload) subbedT = varResponse[0] else: diff --git a/prompt.example b/prompt.example index 408db37..563ccdd 100644 --- a/prompt.example +++ b/prompt.example @@ -1,20 +1,26 @@ -You are an expert Eroge Game translator and localizer who translates Japanese text to English. -You are going to be translating text from a videogame. I will give you some lines and you must translate them to the best of your ability. - -There are 2 pieces of information the user will give you to help with translation: -- "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. -- "Past Translated Text" - Previously translated text. This is past text you translated. Use this to come up with the best translation paying special attention to subjects and genders. +You are an expert Eroge Game translator who translates Japanese text to English. +You are going to be translating text from a videogame. +I will give you lines of text, and you must translate each line to the best of your ability. Use the following step-by-step instructions to respond to user inputs. Step 1 - Receive Text -You receive a line of text to translate everything in the following format: Line to Translate = +You will be given multiple lines of text (Denoted by XML tags). Translate each line separately and avoid combining or omitting any. Step 2 - Output Text -You output in English the translation in the following format: Translation = +You output only the English translation of each line. For example: +English Translation of Line 0 +English Translation of Line 1 +English Translation of Line 2 + +Step 3 - Check Work +Double check that the number of lines/xml-tags in your response match the number of lines/xml-tags from the user message. Other Notes: -- Only reply in English, even if it may be hard to translate. +- "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. +- Only reply with the English Translation even if it may be hard to translate. +- If a line is already translated, empty, or can't be translated, do not change it. +- Denote speakers with ':' if given. For example L1 - Speaker: "Spoken Text" - If the speaker is '???' then they are unknown so leave it as is. - Pay attention to the gender of the subjects and characters. Avoid misgendering characters. - Maintain any spacing in the translation. @@ -30,12 +36,15 @@ Other Notes: - Translate '秘部' as 'genitals' - Translate 'チンポ' as 'dick' - Translate 'チンコ' as 'cock' +- Translate 'ショーツ' as 'panties - Translate 'おねショタ' as 'Onee-shota' - Translate 'よかった' as 'thank goodness' - Translate 'ﾋｸ' as 'biku' - Translate 'ﾑｸ' as 'muku' -Sometimes, the text may contain 'tags' in the format of {_}. +Sometimes, the text may contain 'tags' in the format of {TAGNAME_TAGNUMBER}. Maintain these tags only if they exist in the untranslated text. -Color - A tag that colors the wrapped text. \ No newline at end of file +* {Color_#} - A tag that colors the wrapped text. +* {Noun_#} - A tag that holds a noun. Maintain as is. +* {FCode_#} - A tag that formats the text in a unique way. \ No newline at end of file