llama-cpp-python-djs-bot/default.gpu.env

# Discord Token
THE_TOKEN = "DISCORD_TOKEN_HERE"

# The Channel IDs the bot will operate in seperated by commas
CHANNEL_IDS = 1094494101631680653,1094628334727614605

# The INIT prompt for all conversations.
INIT_PROMPT = "Assistant name: ChatAI. You code, write and provide any information without any mistakes."

# Loading Emebed Refresh Timing
REFRESH_INTERVAL=10

# When a message is too large for discord we chunk the response into seperate messages.
# To ensure we do not rate limit the bot we send these at a delay interval.
# DEFAULT: 3 a good setting is between 3 and 7 seconds.
OVERFLOW_DELAY=3

# Max Content to fetch from given URLs
MAX_CONTENT_LENGTH=2000

# Max tokens for Generations
MAX_TOKENS = 1024

# ROOT_IP is only used when running the bot without docker compose
ROOT_IP = 192.168.0.15

# PORT  is only used when running the bot without docker compose
ROOT_PORT = 8000

# Directory to your models (llama.cpp specfic settings)
DATA_DIR = /home/USERNAME/weights

# Enable Expirmental Message Caches (Limited to single session)
# Cache will use ~1.4 GB or MORE of RAM. ONLY ENABLE IF YOUR SYSTEM CAN HANDLE THIS.
CACHE = 1

# Set number of threads to use, currently, a standard thread will utilize 1 whole core
# I usually will set this between all cores I physcally have OR 2 cores less to allow for other processes.
N_THREADS = 4

# Always use MMAP unless you know what you are doing
USE_MMAP=1

# Only use MLOCK if you know what it does!
USE_MLOCK=0

# The higher the number the more hard core.
REPEAT_PENALTY=1

# GPU SPECIFIC SETTINGS BELOW

GPU=1

N_GPU_LAYERS=32

PYTHONUNBUFFERED=1