Merge branch 'main' of ssh.gitgud.io:DazedAnon/DazedMTLTool

This commit is contained in:
dazedanon 2025-10-24 11:20:22 -05:00
commit 3328d898f0
6 changed files with 343 additions and 161 deletions

View file

@ -1,3 +1,17 @@
# --------------------------------------------------
# Gemini-Specific Settings (Optional)
# --------------------------------------------------
# Sets the thinking budget for Gemini models.
# - To turn on dynamic thinking, set to -1.
# - To disable thinking (on compatible models like 2.5 Flash), set to 0.
# - To set a specific token budget, use a positive integer (e.g., 8192).
# - Leave this variable out to use the model's default setting.
GEMINI_THINKING_BUDGET=
# Set to "gemini" to use the Gemini API or "openai" for OpenAI (If empty it will default to openai.)
API_PROVIDER=openai
#API link, leave blank to use OpenAI API
api=""

View file

@ -239,4 +239,4 @@ def deleteFolderFiles(folderPath):
for filename in os.listdir(folderPath):
file_path = os.path.join(folderPath, filename)
if file_path.endswith((".json", ".yaml", ".ks")):
os.remove(file_path)
os.remove(file_path)

View file

@ -18,9 +18,14 @@ import tempfile
# Open AI
load_dotenv()
if os.getenv("api").replace(" ", "") != "":
openai.base_url = os.getenv("api")
openai.organization = os.getenv("org")
if os.getenv("API_PROVIDER") == "gemini":
openai.base_url = "https://generativelanguage.googleapis.com/v1beta/openai/"
openai.organization = None
else:
api_url = os.getenv("api")
if api_url and api_url.replace(" ", "") != "":
openai.base_url = api_url
openai.organization = os.getenv("org")
openai.api_key = os.getenv("key")
# Globals
@ -429,4 +434,4 @@ def translateAI(text, history, fullPromptFlag):
pbar=PBAR,
lock=LOCK,
mismatchList=MISMATCH
)
)

View file

@ -89,7 +89,7 @@ CODE102 = True
# Optional
CODE101 = False
CODE408 = False
CODE408 = True
# Variables
CODE122 = False
@ -544,15 +544,61 @@ def parseSystem(data, filename):
with tqdm(total=total_units, bar_format=BAR_FORMAT, position=POSITION, leave=LEAVE) as pbar:
pbar.desc = filename
input_log = {}
output_log = {}
try:
# Helper to flatten lists to Line1, Line2, ...
def flatten_lines(prefix, values, target):
for idx, v in enumerate(values):
target[f"{prefix}Line{idx+1}"] = v
# Capture pre-translation values
if "terms" in data:
for key, value in data["terms"].items():
if isinstance(value, list):
flatten_lines(key, value, input_log)
else:
input_log[key] = value
for key in ["gameTitle", "game_title", "armor_types", "skill_types", "weapon_types"]:
if key in data:
val = data[key]
if isinstance(val, list):
flatten_lines(key, val, input_log)
else:
input_log[key] = val
# Run translation
result = searchSystem(data, pbar)
totalTokens[0] += result[0]
totalTokens[1] += result[1]
# Capture post-translation values
if "terms" in data:
for key, value in data["terms"].items():
if isinstance(value, list):
flatten_lines(key, value, output_log)
else:
output_log[key] = value
for key in ["gameTitle", "game_title", "armor_types", "skill_types", "weapon_types"]:
if key in data:
val = data[key]
if isinstance(val, list):
flatten_lines(key, val, output_log)
else:
output_log[key] = val
except Exception as e:
traceback.print_exc()
return [data, totalTokens, e]
finally:
maybe_save_progress_yaml(data, filename, result)
# Log translation history for system.yaml in established line-by-line format
try:
with open("log/translationHistory.txt", "a", encoding="utf-8") as log_file:
log_file.write("Input:\n")
log_file.write(json.dumps(input_log, ensure_ascii=False, indent=4))
log_file.write("\nOutput:\n")
log_file.write(json.dumps(output_log, ensure_ascii=False, indent=4))
log_file.write("\n")
except Exception:
pass
return [data, totalTokens, None]
@ -1091,7 +1137,7 @@ def searchCodes(page, pbar, jobList, filename):
# Brackets
if len(speakerList) == 0:
speakerList = re.findall(r"^【(.*?)】$|^【(.*?)】[\\]*[a-zA-Z]*\[.*\]$", jaString)
speakerList = re.findall(r"^(?:[\\]+[rlRL]\[[\w\d\-]+\])?【(.*?)】$|^(?:[\\]+[rlRL]\[[\w\d\-]+\])?【(.*?)】[\\]*[a-zA-Z]*\[.*\]$", jaString)
if speakerList:
if speakerList[0][0]:
speakerList = [speakerList[0][0]]
@ -1114,9 +1160,9 @@ def searchCodes(page, pbar, jobList, filename):
# First Line Speakers
if len(speakerList) == 0 and FIRSTLINESPEAKERS is True:
# Remove any RPGMaker Code at start
# Remove any RPGMaker Code at start (including \r[...] and \l[...] patterns)
ffMatch = re.search(
r"^((?:[\\]+[^cCnNiIkKvV]+\[[\d\w]+\])+)",
r"^((?:[\\]+[^cCnNiIkKvV]+\[[\d\w]+\])+|(?:[\\]+[rlRL]\[[\w\d\-]+\]))",
jaString,
)
if ffMatch != None:
@ -1152,7 +1198,13 @@ def searchCodes(page, pbar, jobList, filename):
# Set Data
if not setData:
codeList[i]["p"][0] = nametag + jaString.replace(speakerList[0], speaker)
# Check if there's a \r[...] or \l[...] code at the beginning
codePrefix = re.search(r"^([\\]+[rlRL]\[[\w\d\-]+\])", jaString)
if codePrefix:
# Replace only the speaker name, preserving the code and brackets
codeList[i]["p"][0] = jaString.replace(f"{speakerList[0]}", f"{speaker}")
else:
codeList[i]["p"][0] = nametag + jaString.replace(speakerList[0], speaker)
nametag = ""
# Iterate to next string
@ -1761,24 +1813,40 @@ def searchCodes(page, pbar, jobList, filename):
## Event Code: 408 (Script)
if "c" in codeList[i] and (codeList[i]["c"] == 408) and CODE408 is True:
# Remove Textwrap
jaString = codeList[i]["p"][0]
# jaString = jaString.replace("\n", " ")
# Save starting index
j = i
jaString = codeList[i]["p"][0] if len(codeList[i]["p"]) > 0 else ""
# Join consecutive 408 codes
combinedText = jaString
if len(codeList) > i + 1:
while i + 1 < len(codeList) and "c" in codeList[i + 1] and codeList[i + 1]["c"] == 408:
i += 1
if len(codeList[i]["p"]) > 0:
combinedText += " " + codeList[i]["p"][0]
# Mark as -1 only in Pass 2
if not setData:
codeList[i]["p"] = []
codeList[i]["c"] = -1
# Pass 1
if setData:
list408.append(jaString)
if combinedText.strip(): # Only add non-empty text
list408.append(combinedText)
# Pass 2
else:
translatedText = list408[0]
list408.pop(0)
if len(list408) > 0:
translatedText = list408[0]
list408.pop(0)
# Textwrap
translatedText = dazedwrap.wrapText(translatedText, width=1000)
# Textwrap
translatedText = dazedwrap.wrapText(translatedText, width=WIDTH)
# Set Data
codeList[i]["p"][0] = f"{translatedText}"
# Set Data - ensure list has at least one element
if len(codeList[j]["p"]) == 0:
codeList[j]["p"] = [""]
codeList[j]["p"][0] = f"{translatedText}"
## Event Code: 108 (Script)
if "c" in codeList[i] and (codeList[i]["c"] == 108) and CODE108 is True:
@ -2449,15 +2517,23 @@ def searchSystem(data, pbar):
totalTokens = [0, 0]
context = "Reply with only the " + LANGUAGE + ' translation of the UI textbox."'
# Title
response = translateAI(
data["gameTitle"],
" Reply with the " + LANGUAGE + " translation of the game title name",
False,
)
totalTokens[0] += response[1][0]
totalTokens[1] += response[1][1]
data["gameTitle"] = response[0].strip(".")
# Title (handle both 'gameTitle' and 'game_title', skip if missing)
game_title = data.get("gameTitle")
if game_title is None:
game_title = data.get("game_title")
if game_title is not None:
response = translateAI(
game_title,
" Reply with the " + LANGUAGE + " translation of the game title name",
False,
)
totalTokens[0] += response[1][0]
totalTokens[1] += response[1][1]
# Write back to whichever key existed
if "gameTitle" in data:
data["gameTitle"] = response[0].strip(".")
else:
data["game_title"] = response[0].strip(".")
# Terms
for term in data["terms"]:
@ -2510,26 +2586,44 @@ def searchSystem(data, pbar):
# totalTokens[1] += response[1][1]
# data['variables'][i] = response[0].replace('\"', '').strip()
# Messages
# Messages and lists (handle both string and list values, log as Line1, Line2, ...)
messages = data["terms"]
for key, value in messages.items():
response = translateAI(
value,
"Reply with only the "
+ LANGUAGE
+ ' translation of the battle text.\nTranslate "常時ダッシュ" as "Always Dash"\nTranslate "次の%1まで" as Next %1.',
False,
)
translatedText = response[0]
# Remove characters that may break scripts
charList = [".", '"', "\\n"]
for char in charList:
translatedText = translatedText.replace(char, "")
totalTokens[0] += response[1][0]
totalTokens[1] += response[1][1]
messages[key] = translatedText
if isinstance(value, list):
new_list = []
for idx, item in enumerate(value):
if isinstance(item, str):
response = translateAI(
item,
f"Reply with only the {LANGUAGE} translation of the battle text.\nTranslate '常時ダッシュ' as 'Always Dash'\nTranslate '次の%1まで' as Next %1.",
False,
)
translatedText = response[0]
charList = [".", '"', "\\n"]
for char in charList:
translatedText = translatedText.replace(char, "")
totalTokens[0] += response[1][0]
totalTokens[1] += response[1][1]
new_list.append(translatedText)
else:
new_list.append(item)
# For logging: create Line1, Line2, ... keys for this list
messages[key] = new_list
elif isinstance(value, str):
response = translateAI(
value,
f"Reply with only the {LANGUAGE} translation of the battle text.\nTranslate '常時ダッシュ' as 'Always Dash'\nTranslate '次の%1まで' as Next %1.",
False,
)
translatedText = response[0]
charList = [".", '"', "\\n"]
for char in charList:
translatedText = translatedText.replace(char, "")
totalTokens[0] += response[1][0]
totalTokens[1] += response[1][1]
messages[key] = translatedText
else:
messages[key] = value
return totalTokens

View file

@ -18,9 +18,14 @@ import tempfile
# Open AI
load_dotenv()
if os.getenv("api").replace(" ", "") != "":
openai.base_url = os.getenv("api")
openai.organization = os.getenv("org")
if os.getenv("API_PROVIDER") == "gemini":
openai.base_url = "https://generativelanguage.googleapis.com/v1beta/openai/"
openai.organization = None
else:
api_url = os.getenv("api")
if api_url and api_url.replace(" ", "") != "":
openai.base_url = api_url
openai.organization = os.getenv("org")
openai.api_key = os.getenv("key")
# Globals
@ -370,4 +375,4 @@ def translateAI(text, history, fullPromptFlag):
pbar=PBAR,
lock=LOCK,
mismatchList=MISMATCH
)
)

View file

@ -130,6 +130,41 @@ def getPricingConfig(model):
"batchSize": 30,
"frequencyPenalty": 0.05
}
elif "gemini-2.0-flash-lite" in model:
return {
"inputAPICost": 0.075,
"outputAPICost": 0.30,
"batchSize": 30,
"frequencyPenalty": 0.0
}
elif "gemini-2.0-flash" in model:
return {
"inputAPICost": 0.10,
"outputAPICost": 0.40,
"batchSize": 30,
"frequencyPenalty": 0.0
}
elif "gemini-2.5-flash-lite" in model:
return {
"inputAPICost": 0.10,
"outputAPICost": 0.40,
"batchSize": 30,
"frequencyPenalty": 0.0
}
elif "gemini-2.5-flash" in model:
return {
"inputAPICost": 0.30,
"outputAPICost": 2.50,
"batchSize": 30,
"frequencyPenalty": 0.0
}
elif "gemini-2.5-pro" in model:
return {
"inputAPICost": 1.25,
"outputAPICost": 10.00,
"batchSize": 30,
"frequencyPenalty": 0.0
}
else:
# Fallback to environment variables
return {
@ -285,7 +320,7 @@ def createTranslationSchema(numLines):
def translateText(system, user, history, penalty, formatType, model, numLines=None):
"""Send translation request to OpenAI API"""
"""Send translation request to the selected API"""
# Ensure system content is not empty
if not system or not str(system).strip():
raise ValueError("System content cannot be empty")
@ -330,42 +365,53 @@ def translateText(system, user, history, penalty, formatType, model, numLines=No
if not message.get("content") or not str(message.get("content")).strip():
raise ValueError(f"Message {i} has empty content: {message}")
# Call OpenAI API
try:
# --- API Call Logic ---
api_provider = os.getenv("API_PROVIDER", "openai").lower()
# Base parameters for the API call
params = {
"model": model,
"response_format": responseFormat,
"messages": msg,
}
# Provider-specific parameters
if api_provider == "gemini":
params["temperature"] = 0
# Handle thinking budget for Gemini
thinking_budget_str = os.getenv("GEMINI_THINKING_BUDGET")
if thinking_budget_str:
try:
thinking_budget = int(thinking_budget_str)
params["extra_body"] = {
'google': {
'thinking_config': {
'thinking_budget': thinking_budget
}
}
}
except (ValueError, TypeError):
# Ignore if the value is not a valid integer
pass
# frequency_penalty is not supported via the OpenAI compatibility layer for Gemini
else: # Default to OpenAI behavior
if "gpt-5" in model:
response = openai.chat.completions.create(
model=model,
response_format=responseFormat,
messages=msg,
reasoning_effort="minimal"
)
params["reasoning_effort"] = "minimal"
else:
response = openai.chat.completions.create(
model=model,
response_format=responseFormat,
messages=msg,
temperature=0,
frequency_penalty=penalty
)
params["temperature"] = 0
params["frequency_penalty"] = penalty
# Call API
try:
response = openai.chat.completions.create(**params)
except Exception as e:
# If structured output fails, fallback to json_object
if formatType == "json" and "json_schema" in str(responseFormat):
responseFormat = {"type": "json_object"}
if "gpt-5" in model:
response = openai.chat.completions.create(
model=model,
response_format=responseFormat,
messages=msg,
reasoning_effort="minimal"
)
else:
response = openai.chat.completions.create(
model=model,
response_format=responseFormat,
messages=msg,
temperature=0,
frequency_penalty=penalty
)
params["response_format"] = responseFormat
response = openai.chat.completions.create(**params)
else:
raise e
@ -567,7 +613,6 @@ def translateAI(text, history, fullPromptFlag, config, filename=None, pbar=None,
return [text, [0, 0]]
with open(config.logFilePath, "a+", encoding="utf-8") as logFile:
mismatch = False
totalTokens = [0, 0]
if isinstance(text, list):
@ -612,97 +657,116 @@ def translateAI(text, history, fullPromptFlag, config, filename=None, pbar=None,
totalTokens[1] += estimate[1]
continue
# Translate
# --- Translation and Validation Retry Block ---
max_retries = 2 # 1 initial attempt + 2 retries
final_translations = None
last_raw_translation = ""
numLines = len(tItem) if isinstance(tItem, list) else None
response = translateText(system, user, history, 0.05, formatType, config.model, numLines)
translatedText = response.choices[0].message.content
# Retry if AI refused
if not translatedText or '"error":' in translatedText:
response = translateText(
f"{system}\n You translate ALL content.",
user, history, 0.1, formatType, "gpt-4.1" if config.model == "gpt-5" else "gpt-4o", numLines
)
for attempt in range(max_retries + 1):
is_valid = True
# On retries, add a note to the system prompt
current_system = system
if attempt > 0:
current_system += f"\n\nIMPORTANT: Your previous attempt was incorrect or incomplete. Please ensure the entire output is translated to {config.language} and contains no untranslated characters. Translate the following text again, ensuring the JSON structure is correct."
if pbar:
pbar.write(f"Retrying translation... (Attempt {attempt + 1}/{max_retries + 1})")
# Translate
response = translateText(current_system, user, history, 0.05, formatType, config.model, numLines)
translatedText = response.choices[0].message.content
last_raw_translation = translatedText
# Update token count
totalTokens[0] += response.usage.prompt_tokens
totalTokens[1] += response.usage.completion_tokens
# Update token count for this attempt
totalTokens[0] += response.usage.prompt_tokens
totalTokens[1] += response.usage.completion_tokens
# Clean the translation first for consistency
translatedText = cleanTranslatedText(translatedText, config.language)
# Clean the translation first for consistency
cleaned_text = cleanTranslatedText(translatedText, config.language)
# Process translation result
if translatedText:
if isinstance(tItem, list):
extractedTranslations = extractTranslation(translatedText, True, pbar)
if extractedTranslations is None or len(tItem) != len(extractedTranslations):
# Mismatch, try again
response = translateText(system, user, history, 0.05, formatType, config.model, numLines)
translatedText = response.choices[0].message.content
totalTokens[0] += response.usage.prompt_tokens
totalTokens[1] += response.usage.completion_tokens
# Clean and extract again
translatedText = cleanTranslatedText(translatedText, config.language)
extractedTranslations = extractTranslation(translatedText, True, pbar)
if extractedTranslations is None or len(tItem) != len(extractedTranslations):
# Format the JSON output for consistent mismatch logging
formatted_mismatch_output = translatedText
try:
parsed_json = json.loads(translatedText)
formatted_mismatch_output = json.dumps(parsed_json, indent=4, ensure_ascii=False)
except (json.JSONDecodeError, ValueError):
pass
# Log mismatch
with open(config.mismatchLogPath, "a+", encoding="utf-8") as mismatchFile:
mismatchFile.write(f"Mismatch: {filename}\n")
mismatchFile.write(f"Input:\n{subbedT}\n")
mismatchFile.write(f"Output:\n{formatted_mismatch_output}\n")
mismatch = True
# Format the JSON output for consistent logging
formatted_output = translatedText
try:
# Try to parse and reformat the JSON for consistent formatting
parsed_json = json.loads(translatedText)
formatted_output = json.dumps(parsed_json, indent=4, ensure_ascii=False)
except (json.JSONDecodeError, ValueError):
# If it's not valid JSON, keep the original
pass
logFile.write(f"Input:\n{subbedT}\n")
logFile.write(f"Output:\n{formatted_output}\n")
# Set results if no mismatch
if not mismatch:
tList[index] = extractedTranslations
history = extractedTranslations[-config.maxHistory:]
# Process and validate translation result
if cleaned_text:
if isinstance(tItem, list):
extracted = extractTranslation(cleaned_text, True, pbar)
# Check 1: Mismatch in length
if extracted is None or len(tItem) != len(extracted):
is_valid = False
# Check 2: Untranslated content
else:
for line in extracted:
if re.search(config.langRegex, str(line)):
is_valid = False
break
if is_valid:
final_translations = extracted
else:
history = text[-config.maxHistory:] if isinstance(text, list) else text
mismatch = False
if filename and mismatchList is not None and filename not in mismatchList:
mismatchList.append(filename)
# Update progress bar
if lock and pbar is not None:
with lock:
pbar.update(len(tItem))
# Check for untranslated content in single string
if re.search(config.langRegex, cleaned_text):
is_valid = False
else:
final_translations = cleaned_text.replace("Placeholder Text", "")
else:
# Single string translation - clean after getting the raw response
cleanedText = cleanTranslatedText(translatedText, config.language)
tList[index] = cleanedText.replace("Placeholder Text", "")
else:
if pbar:
pbar.write(f"AI Refused: {tItem}\n")
is_valid = False
if pbar: pbar.write(f"AI Refused: {tItem}\n")
# If translation is valid, break the retry loop
if is_valid:
break
# --- End of Retry Block ---
# After the loop, handle the final result
if final_translations is not None: # Success case
formatted_output = last_raw_translation
try:
parsed_json = json.loads(last_raw_translation)
formatted_output = json.dumps(parsed_json, indent=4, ensure_ascii=False)
except (json.JSONDecodeError, ValueError):
pass
logFile.write(f"Input:\n{subbedT}\n")
logFile.write(f"Output:\n{formatted_output}\n")
if isinstance(tItem, list):
tList[index] = final_translations
history = final_translations[-config.maxHistory:]
else:
tList[index] = final_translations
history = final_translations
if lock and pbar is not None:
with lock:
pbar.update(len(tItem) if isinstance(tItem, list) else 1)
else: # Failure case after all retries
if pbar: pbar.write(f"Translation failed after {max_retries + 1} attempts. Check mismatch log.")
formatted_mismatch_output = last_raw_translation
try:
parsed_json = json.loads(last_raw_translation)
formatted_mismatch_output = json.dumps(parsed_json, indent=4, ensure_ascii=False)
except (json.JSONDecodeError, ValueError):
pass
with open(config.mismatchLogPath, "a+", encoding="utf-8") as mismatchFile:
mismatchFile.write(f"Failed after retries: {filename}\n")
mismatchFile.write(f"Input:\n{subbedT}\n")
mismatchFile.write(f"Final Output:\n{formatted_mismatch_output}\n")
if filename and mismatchList is not None and filename not in mismatchList:
mismatchList.append(filename)
tList[index] = tItem
history = text[-config.maxHistory:] if isinstance(text, list) else text
# Combine if multilist
if isinstance(tList[0], list):
if tList and isinstance(tList[0], list):
tList = [t for sublist in tList for t in sublist]
# Return result
if formatType == "json":
return [tList, totalTokens]
else:
return [tList[0], totalTokens]
return [tList[0], totalTokens]