core/settings.py

import os
from dataclasses import dataclass
from pathlib import Path

from dotenv import load_dotenv

DEFAULT_LLM_MODEL_FILE = "Llama-3.2-1B-Instruct-Q4_K_M.gguf"


def read_int(name, default_value):
    raw_value = os.getenv(name, str(default_value)).strip()
    try:
        return int(raw_value)
    except ValueError:
        return default_value


def resolve_path(root_dir, raw_value):
    path = Path(raw_value)
    if path.is_absolute():
        return path
    return root_dir / path


def resolve_llm_model_path(models_dir, model_file):
    if model_file:
        requested_path = models_dir / model_file
        if requested_path.exists():
            return requested_path

    default_path = models_dir / DEFAULT_LLM_MODEL_FILE
    if default_path.exists():
        return default_path

    gguf_files = sorted(models_dir.glob("*.gguf"))
    if gguf_files:
        return gguf_files[0]

    return models_dir / (model_file or DEFAULT_LLM_MODEL_FILE)


def make_local_model_dir(hf_cache_dir, repo_id):
    safe_name = repo_id.replace("/", "--")
    return hf_cache_dir / "downloads" / safe_name


def resolve_hf_source(hf_cache_dir, repo_id):
    local_dir = make_local_model_dir(hf_cache_dir, repo_id)
    if local_dir.exists():
        return local_dir
    return repo_id


@dataclass(frozen=True)
class Settings:
    project_root: Path
    models_dir: Path
    hf_cache_dir: Path
    data_dir: Path
    llm_model_file: str
    llm_model_path: Path
    context_size: int
    live_asr_model_id: str
    live_asr_source: str | Path
    batch_asr_model_id: str
    batch_asr_source: str | Path
    image_model_id: str
    image_model_source: str | Path
    tts_en_model_id: str
    tts_en_source: str | Path
    tts_hi_model_id: str
    tts_hi_source: str | Path
    asr_compute_type: str
    embedding_model_id: str
    tesseract_cmd: str


def load_settings():
    project_root = Path(__file__).resolve().parent.parent
    load_dotenv(project_root / ".env")

    models_dir = resolve_path(project_root, os.getenv("EDUAI_MODELS_DIR", "models"))
    hf_cache_dir = resolve_path(project_root, os.getenv("EDUAI_HF_CACHE_DIR", "models/hf_cache"))
    data_dir = resolve_path(project_root, os.getenv("EDUAI_DATA_DIR", "data"))

    models_dir.mkdir(parents=True, exist_ok=True)
    hf_cache_dir.mkdir(parents=True, exist_ok=True)
    data_dir.mkdir(parents=True, exist_ok=True)

    os.environ.setdefault("HF_HOME", str(hf_cache_dir))
    os.environ.setdefault("HF_HUB_CACHE", str(hf_cache_dir / "hub"))
    os.environ.setdefault("TOKENIZERS_PARALLELISM", "false")

    llm_model_file = os.getenv("EDUAI_LLM_MODEL_FILE", DEFAULT_LLM_MODEL_FILE).strip()

    live_asr_model_id = os.getenv("EDUAI_LIVE_ASR_MODEL_ID", "Systran/faster-whisper-small").strip()
    batch_asr_model_id = os.getenv("EDUAI_BATCH_ASR_MODEL_ID", "Systran/faster-whisper-large-v3").strip()
    image_model_id = os.getenv("EDUAI_IMAGE_MODEL_ID", "HuggingFaceTB/SmolVLM-256M-Instruct").strip()
    tts_en_model_id = os.getenv("EDUAI_TTS_EN_MODEL_ID", "facebook/mms-tts-eng").strip()
    tts_hi_model_id = os.getenv("EDUAI_TTS_HI_MODEL_ID", "facebook/mms-tts-hin").strip()
    embedding_model_id = os.getenv("EDUAI_EMBEDDING_MODEL_ID", "sentence-transformers/all-MiniLM-L6-v2").strip()

    return Settings(
        project_root=project_root,
        models_dir=models_dir,
        hf_cache_dir=hf_cache_dir,
        data_dir=data_dir,
        llm_model_file=llm_model_file,
        llm_model_path=resolve_llm_model_path(models_dir, llm_model_file),
        context_size=read_int("EDUAI_CONTEXT_SIZE", 8192),
        live_asr_model_id=live_asr_model_id,
        live_asr_source=resolve_hf_source(hf_cache_dir, live_asr_model_id),
        batch_asr_model_id=batch_asr_model_id,
        batch_asr_source=resolve_hf_source(hf_cache_dir, batch_asr_model_id),
        image_model_id=image_model_id,
        image_model_source=resolve_hf_source(hf_cache_dir, image_model_id),
        tts_en_model_id=tts_en_model_id,
        tts_en_source=resolve_hf_source(hf_cache_dir, tts_en_model_id),
        tts_hi_model_id=tts_hi_model_id,
        tts_hi_source=resolve_hf_source(hf_cache_dir, tts_hi_model_id),
        asr_compute_type=os.getenv("EDUAI_ASR_COMPUTE_TYPE", "int8").strip() or "int8",
        embedding_model_id=embedding_model_id,
        tesseract_cmd=os.getenv("TESSERACT_CMD", "").strip(),
    )