config.py

"""Allowlist, caps and parse prompt — edit models here, not in adapters."""

from __future__ import annotations

import os
from dataclasses import dataclass
from pathlib import Path

from dotenv import load_dotenv

# Repo root: backend/app/config.py -> ../../
_ROOT = Path(__file__).resolve().parents[2]
load_dotenv(_ROOT / ".env")

HF_TOKEN: str = os.getenv("HF_TOKEN", "")
MAX_PAGES: int = 8
DEFAULT_DPI: int = 150
PAGE_TIMEOUT_S: int = 60
CORS_ORIGINS: list[str] = [
    "http://localhost:5173",
    "http://127.0.0.1:5173",
]

PARSE_PROMPT_VERSION = "v1"
PARSE_PROMPT = """Parse this document page into Markdown.
Rules:
- Preserve reading order (top-to-bottom, left-to-right for multi-column).
- Represent tables as GitHub-flavored Markdown pipe tables.
- Use # / ## / ### for section headings found in the document.
- Keep numbers as written in the source (do not convert currency).
- Do not invent content that is not visible.
- Output only Markdown, no preamble."""


@dataclass(frozen=True)
class ModelSpec:
    id: str
    label: str


# Seed allowlist — verified with current HF provider entitlements.
# google/gemma-3-4b-it returns model_not_supported for this account now.
MODELS: list[ModelSpec] = [
    ModelSpec(id="google/gemma-3-12b-it", label="Gemma 3 12B IT"),
    ModelSpec(id="google/gemma-4-26B-A4B-it", label="Gemma 4 26B A4B IT"),
]

DEFAULT_MODEL_ID: str = MODELS[0].id


def model_ids() -> set[str]:
    return {m.id for m in MODELS}


def get_model(model_id: str) -> ModelSpec | None:
    for m in MODELS:
        if m.id == model_id:
            return m
    return None