from config.model import MODEL_DIR

MODEL = MODEL_DIR / "granite-4.2" / "granite-4.2-8b-Q4_K_M.gguf"

N_THREADS      = 10
N_CTX          = 8192  # Context window (tokens)
N_GPU_LAYERS   = 0     # 0 = full CPU, >0 offload N layers to GPU
MAX_TOKENS     = 1024  # Max tokens per reply
STREAM         = True  # Stream tokens as they are generated

TEMPERATURE = 0.7
TOP_P       = 0.9
TOP_K       = 40

# System prompt for the chat session.
SYSTEM_PROMPT = "You are a helpful, concise assistant."

# Default chat template. Leave empty to use the template embedded in the GGUF.
CHAT_TEMPLATE = ""