102 lines
5.6 KiB
Python
102 lines
5.6 KiB
Python
import os, ast, torch, platform
|
|
from dotenv import load_dotenv
|
|
load_dotenv(override=True)
|
|
|
|
if platform.system() == 'Windows':
|
|
default_tmp_dir = "C:\\Users\\AppData\\Local\\Temp"
|
|
elif platform.system() in ['Linux', 'Darwin']:
|
|
default_tmp_dir = "/tmp"
|
|
|
|
|
|
|
|
###################################################################################################
|
|
### EDIT THESE VARIABLES ###
|
|
# Create a .env file in the same directory as this file and add the variables there. Of course you can choose to edit this file but then if you pull from the repository again all the config will be goneeee unless perhaps it is saved or stashed.
|
|
# The default values should be fine for most cases. Only ones that you need to change are the API keys, and the variables under Translation and API Translation if you choose to use an external API.
|
|
# available languages: 'ch_sim', 'ch_tra', 'ja', 'ko', 'en'
|
|
|
|
INTERVAL = float(os.getenv('INTERVAL', 1.5)) # Interval in seconds between translations. If your system is slow, a lower value is probably fine with regards to API rates.
|
|
|
|
### OCR
|
|
IMAGE_CHANGE_THRESHOLD = float(os.getenv('IMAGE_CHANGE_THRESHOLD', 0.75)) # higher values mean more sensitivity to changes in the screen, too high and the screen will constantly refresh
|
|
OCR_MODEL = os.getenv('OCR_MODEL', 'easy') # 'easy', 'paddle', 'rapid' ### easy is the most accurate, paddle is the fastest with CUDA and rapid is the fastest with CPU. Rapid has only between Chinese and English unless you add more languages
|
|
OCR_USE_GPU = ast.literal_eval(os.getenv('OCR_USE_GPU', 'True')) # True or False to use CUDA for OCR. Defaults to CPU if no CUDA GPU is available
|
|
|
|
|
|
|
|
|
|
### Drawing/Overlay Config
|
|
FILL_COLOUR = os.getenv('FILL_COLOUR', 'white') # colour of the textboxes
|
|
FONT_COLOUR = os.getenv('FONT_COLOUR', "#ff0000") # colour of the font
|
|
FONT_FILE = os.getenv('FONT_FILE', os.path.join(__file__, "fonts", "Arial-Unicode-Bold.ttf")) # path to the font file. Ensure it is a unicode .ttf file if you want to be able to see most languages.
|
|
FONT_SIZE_MAX = int(os.getenv('FONT_SIZE_MAX', 20)) # Maximum font size you want to be able to see onscreen
|
|
FONT_SIZE_MIN = int(os.getenv('FONT_SIZE_MIN', 8)) # Minimum font size you want to be able to see onscreen
|
|
LINE_SPACING = int(os.getenv('LINE_SPACING', 3)) # spacing between lines of text with the learn modes in DRAW_TRANSLATIONS_MODE
|
|
REGION = ast.literal_eval(os.getenv('REGION','(0,0,2560,1440)')) # (x1, y1, x2, y2) - the region of the screen to capture
|
|
DRAW_TRANSLATIONS_MODE = os.getenv('DRAW_TRANSLATIONS_MODE', 'learn_cover')
|
|
"""
|
|
DRAW_TRANSLATIONS_MODE possible options:
|
|
`learn': adds translated text, original text (should be added so when texts get moved around the translation of which it references is understood) and (optionally with the other TO_ROMANIZE option) romanized text above the original text. Texts can overlap if squished into a corner. Works well for games where texts are sparser
|
|
'learn_cover': same as above but covers the original text with the translated text. Can help with readability and is less cluttered but with sufficiently dense text the texts can still overlap
|
|
'translation_only_cover': cover the original text with the translated text - will not show the original text at all but also will not be affected by overlapping texts
|
|
"""
|
|
|
|
|
|
|
|
|
|
|
|
### Translation
|
|
MAX_TRANSLATE = int(os.getenv('MAX_TRANSLATION', 200)) # Maximum number of phrases to send to the translation model to translate
|
|
SOURCE_LANG = os.getenv('SOURCE_LANG', 'ch_sim') # Translate from 'ch_sim', 'ch_tra', 'ja', 'ko', 'en'
|
|
TARGET_LANG = os.getenv('TARGET_LANG', 'en') # Translate to 'ch_sim', 'ch_tra', 'ja', 'ko', 'en'
|
|
TO_ROMANIZE = ast.literal_eval(os.getenv('TO_ROMANIZE', 'True')) # romanize the text or not. Only available for one of the learn modes in DRAW_TRANSLATIONS_MODE. It is added above the original text
|
|
|
|
### API Translation (could be external or a local API)
|
|
# API KEYS
|
|
GEMINI_API_KEY = os.getenv('GEMINI_API_KEY') # https://ai.google.dev/
|
|
GROQ_API_KEY = os.getenv('GROQ_API_KEY') # https://console.groq.com/keys
|
|
# MISTRAL_API_KEY = os.getenv('MISTRAL_API_KEY') # https://console.mistral.ai/api-keys/ slow asf
|
|
|
|
### Local Translation Models
|
|
TRANSLATION_MODEL= os.environ['TRANSLATION_MODEL'] # 'opus' or 'm2m' # opus is a lot more lightweight
|
|
TRANSLATION_USE_GPU = ast.literal_eval(os.getenv('TRANSLATION_USE_GPU', 'True'))
|
|
MAX_INPUT_TOKENS = int(os.getenv('MAX_INPUT_TOKENS', 512))
|
|
MAX_OUTPUT_TOKENS = int(os.getenv('MAX_OUTPUT_TOKENS', 512))
|
|
BATCH_SIZE = int(os.getenv('BATCH_SIZE', 6))
|
|
LOCAL_FILES_ONLY = ast.literal_eval(os.getenv('LOCAL_FILES_ONLY', 'False')) # will not attempt pinging Huggingface for the models and just use the cached local models
|
|
|
|
|
|
|
|
###################################################################################################
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
###################################################################################################
|
|
### DO NOT EDIT THESE VARIABLES ###
|
|
## Filepaths
|
|
API_MODELS_FILEPATH = os.path.join(os.path.dirname(__file__), 'api_models.json')
|
|
|
|
|
|
FONT_SIZE = int((FONT_SIZE_MAX + FONT_SIZE_MIN)/2)
|
|
LINE_HEIGHT = FONT_SIZE
|
|
|
|
if TRANSLATION_USE_GPU is False:
|
|
device = torch.device("cpu")
|
|
else:
|
|
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
|
|
|
TEMP_IMG_DIR = os.getenv('TEMP_IMG_PATH', default_tmp_dir) # where the temporary images are stored
|
|
TEMP_IMG_PATH = os.path.join(TEMP_IMG_DIR, 'tempP_img91258102.png')
|
|
### Just for info
|
|
|
|
available_langs = ['ch_sim', 'ch_tra', 'ja', 'ko', 'en'] # there are limitations with the languages that can be used with the OCR models
|
|
seq_llm_models = ['opus', 'm2m']
|
|
api_llm_models = ['gemini']
|
|
causal_llm_models = []
|
|
curr_models = seq_llm_models + api_llm_models + causal_llm_models
|