free_config.py

"""
Free-tier configuration for Week 8 — The Price is Right.

Set GROQ_API_KEY in .env for cloud LLM calls (local run + Hugging Face Spaces).
Set USE_MODAL_SPECIALIST=true only if you want the GPU fine-tuned model on Modal.
By default Modal is OFF — Specialist uses Groq instead (free).
"""

import os

from env_utils import load_project_env

load_project_env()

GROQ_MODEL = os.getenv("GROQ_MODEL", "llama-3.1-8b-instant")
LITELLM_GROQ_MODEL = f"groq/{GROQ_MODEL}"

# Default false: Modal free credits are often only ~$1/month — not enough for regular GPU use.
USE_MODAL_SPECIALIST = os.getenv("USE_MODAL_SPECIALIST", "false").lower() in ("true", "1", "yes")

# Groq when an API key is present; otherwise local Ollama (local dev only).
PREPROCESSOR_MODEL = os.getenv(
    "PRICER_PREPROCESSOR_MODEL",
    LITELLM_GROQ_MODEL if os.getenv("GROQ_API_KEY") else "ollama/llama3.2",
)

MESSAGING_MODEL = os.getenv("MESSAGING_MODEL", LITELLM_GROQ_MODEL)

# Ensemble weights when Modal specialist is unavailable.
WEIGHTS_WITH_SPECIALIST = (0.8, 0.1, 0.1)  # frontier, specialist, neural_network
WEIGHTS_WITHOUT_SPECIALIST = (0.85, 0.15)  # frontier, neural_network

PUSHOVER_CONFIGURED = bool(
    os.getenv("PUSHOVER_USER")
    and os.getenv("PUSHOVER_TOKEN")
    and not os.getenv("PUSHOVER_USER", "").startswith("your-")
)