config.py

max_len = 768 # recommended 512 or 768
sortby="Sno" # ["Sno", "Model", "Device", "Prediction", "Execution Time"]

random_sel = 1 # -1 for all available, else number of random samples to select

# more than 1 will also increase memory usage, adjust accordingly..
num_parallel = 2 # 1 for sequential, 2 or more for parallel


# with all set to GPU mdoe, it is recommended to have atleast 12GB GPU memory
# adjust batch size accordingly to avoid OOM
# if no GPU available, set all to CPU
paths_weights=[
        ('./models/debertav3large', './weights/17_ft1/weights_ep0',   
         1, "gpu"),
        ('./models/debertav3large', './weights/17_ft103/weights_ep0', 
         1, "gpu"),
        ('./models/debertav3large', './weights/19_ft1/weights_ep0',   
         1, "cpu"),
        ('./models/debertav3large', './weights/19_ft103/weights_ep0', 
         1, "cpu"),
        ('./models/debertalarge',   './weights/20_ft1/weights_ep0',   
         1, "gpu"),
        ('./models/debertalarge',   './weights/20_ft103/weights_ep0', 
         1, "cpu"),
    ]


sentiment_weight = 1.0
profanity_threshold = 0.77
profanity_weight_max = 1.1
profanity_weight_min = 0.5
emoji_weight = 0.8
specialChar_weight = 0.2
url_weight = 1.0

zero_shot_thresh = 0.7
zero_shot = True # pre model processing/checks

zero_shot_post = True

check_profanity = True
check_sentiment = True
check_text_score = True

# result_text_thresh = [0.5, 0.27, 0.4] # AI | Human | Likely Human | Updataed to work on memory based chat
result_text_thresh = [0.7, 0.27, 0.55] # AI | Human | Likely Human

debug_mode = False
use_reloader = False
log_output = True
host="0.0.0.0"
port=5000

user_session_timeout = 5 # in minutes

local_mode = False # True for local mode, False for server mode | local mode for using in console, jupyter etc