"""Configuration for the Generic vs Semantic classifier project."""

# Paths
DATA_DIR = "data"
RAW_DIR = f"{DATA_DIR}/raw"
PROCESSED_DIR = f"{DATA_DIR}/processed"
MODELS_DIR = "models"
SCRIPTS_DIR = "scripts"

# Dataset
TOTAL_PER_CATEGORY = 3000
BATCH_SIZE_GEN = 50  # examples per Ollama API call
MAX_CONCURRENT = 8   # parallel API requests

# Categories
CATEGORIES = {
    "en_generic": {"lang": "English", "label": "GENERIC"},
    "en_semantic": {"lang": "English", "label": "SEMANTIC"},
    "hi_generic": {"lang": "Hindi", "label": "GENERIC"},
    "hi_semantic": {"lang": "Hindi", "label": "SEMANTIC"},
}

# Model
MODEL_NAME = "distilbert-base-multilingual-cased"
MAX_SEQ_LEN = 64
NUM_LABELS = 2
LABEL_MAP = {"GENERIC": 0, "SEMANTIC": 1}

# Training
BATCH_SIZE = 32
LEARNING_RATE = 2e-5
NUM_EPOCHS = 5
TEST_SPLIT = 0.15

# Ollama
OLLAMA_URL = "http://localhost:11434/api/chat"
OLLAMA_MODEL = "llama3.1:8b-instruct-q4_K_M"