scripts/_smoke_text_model.py

import argparse
import json
import sys
from pathlib import Path

sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
from config import Settings
from llm.omni_client import OmniClient, _is_reasoning_model

parser = argparse.ArgumentParser()
parser.add_argument("--model-path", required=True)
parser.add_argument("--model-name", default="")
parser.add_argument("--base-url", default="http://127.0.0.1:8080/v1")
args = parser.parse_args()

model_path = Path(args.model_path)
model_name = args.model_name or model_path.name

settings = Settings(
    llm_provider="llama_cpp_server",
    llamacpp_model_path=model_path,
    llm_model=model_name,
    llamacpp_base_url=args.base_url,
    witness_chat_tts=False,
)
client = OmniClient(settings)

print("model:", model_name)
print("reasoning-gated (/no_think injected):", _is_reasoning_model(model_name, model_path))
print("health ready:", client.health().get("ready"))

r = client.chat(
    "You are a terse witness in an English detective game. Answer in one short English sentence.",
    "What color was the folder the suspect carried?",
    task="interview", temperature=0.2,
)
print("INTERVIEW:", repr(r.text))

try:
    d = client.json_chat(
        "You output only a JSON object with keys: summary (string), confidence (number 0-1).",
        "Summarize this sighting: a man in a grey raincoat boarded a bus at 9pm carrying a blue folder.",
        task="witness",
    )
    print("JSON OK:", json.dumps(d)[:300])
except Exception as e:
    print("JSON ERR:", type(e).__name__, e)