services/model_router.py

from services.longcat_client import generate as longcat_generate
from services.model_loader import generate_text as qwen_generate


def _log(msg: str):
    print(f"[Router] {msg}")


def generate_text(messages, temperature=0.3, max_new_tokens=2000):
    text, meta = longcat_generate(messages, temperature, max_new_tokens)

    if text:
        _log(f"LongCat served — model={meta.get('model','?')} tokens={meta.get('tokens','?')}")
        return text

    err = meta.get("error", "unknown")
    if err == "rate_limit":
        _log(f"LongCat quota exhausted — falling back to Qwen")
    elif err == "auth":
        _log(f"LongCat auth failed — falling back to Qwen")
    elif err == "cooldown":
        _log(f"LongCat in cooldown — falling back to Qwen")
    else:
        _log(f"LongCat error ({err}) — falling back to Qwen")

    try:
        text = qwen_generate(messages, temperature, max_new_tokens)
        if text:
            _log("Qwen served (fallback)")
            return text
        _log("Qwen returned None — model may not be loaded")
        return None
    except Exception as e:
        _log(f"Qwen generation failed: {e}")
        return None