| """Smoke test of tool calling with the Meta Model API (Responses API).
|
|
|
| Sends one question that needs ONE dummy tool (``get_beam_width``) at two
|
| positions, runs the returned function calls locally, sends the results back
|
| with ``function_call_output`` and ``previous_response_id``, and prints every
|
| output item, the token usage and the latency of each model call. The tool
|
| returns a made-up number; no real or lab data is sent.
|
|
|
| Usage (from the repository root, PowerShell)::
|
|
|
| .venv\\Scripts\\python.exe scripts\\smoke_test_tools.py
|
| """
|
|
|
| from __future__ import annotations
|
|
|
| import argparse
|
| import json
|
| import os
|
| import sys
|
| import time
|
| from pathlib import Path
|
|
|
| from dotenv import load_dotenv
|
| from openai import OpenAI
|
|
|
| BASE_URL = "https://api.meta.ai/v1"
|
| REPO_ROOT = Path(__file__).resolve().parents[1]
|
| MAX_CALLS = 4
|
|
|
| INSTRUCTIONS = (
|
| "You are testing a tool interface. Use the tool to answer. Always fill in the "
|
| "'reason' argument with a short reason. Answer in one sentence at the end."
|
| )
|
| TOOL = {
|
| "type": "function",
|
| "name": "get_beam_width",
|
| "description": "Dummy tool: returns a beam width in um at a stage position in mm.",
|
| "strict": True,
|
| "parameters": {
|
| "type": "object",
|
| "properties": {
|
| "z_mm": {"type": "number", "description": "Stage position in mm."},
|
| "reason": {"type": "string", "description": "Why you call this tool."},
|
| },
|
| "required": ["z_mm", "reason"],
|
| "additionalProperties": False,
|
| },
|
| }
|
|
|
|
|
| def get_beam_width(z_mm: float, reason: str) -> dict:
|
| """Made-up width: 200 um at 120 mm, growing linearly away from it."""
|
| return {"z_mm": z_mm, "d_um": round(200.0 + 8.0 * abs(z_mm - 120.0), 1)}
|
|
|
|
|
| def main() -> int:
|
| parser = argparse.ArgumentParser(description=__doc__.splitlines()[0])
|
| parser.add_argument("--model", default="muse-spark-1.2")
|
| parser.add_argument("--effort", default="low")
|
| args = parser.parse_args()
|
|
|
| load_dotenv(REPO_ROOT / ".env")
|
| api_key = os.environ.get("MODEL_API_KEY")
|
| if not api_key:
|
| print("MODEL_API_KEY is not set (see .env.example).", file=sys.stderr)
|
| return 1
|
| client = OpenAI(api_key=api_key, base_url=BASE_URL)
|
|
|
| next_input: list | str = (
|
| "What are the beam widths at z = 100 mm and z = 140 mm? Use the tool."
|
| )
|
| previous_id = None
|
| for call in range(1, MAX_CALLS + 1):
|
| start = time.perf_counter()
|
| response = client.responses.create(
|
| model=args.model,
|
| instructions=INSTRUCTIONS,
|
| input=next_input,
|
| tools=[TOOL],
|
| reasoning={"effort": args.effort},
|
| max_output_tokens=4000,
|
| previous_response_id=previous_id,
|
| )
|
| latency = time.perf_counter() - start
|
| usage = response.usage
|
| print(f"\n== model call {call}: status {response.status}, {latency:.2f} s")
|
| print(
|
| f" tokens: input {usage.input_tokens} "
|
| f"(cached {usage.input_tokens_details.cached_tokens}), "
|
| f"output {usage.output_tokens} "
|
| f"(reasoning {usage.output_tokens_details.reasoning_tokens})"
|
| )
|
| outputs = []
|
| for item in response.output:
|
| if item.type == "function_call":
|
| print(
|
| f" function_call: name={item.name} call_id={item.call_id} "
|
| f"arguments={item.arguments}"
|
| )
|
| result = get_beam_width(**json.loads(item.arguments))
|
| outputs.append(
|
| {
|
| "type": "function_call_output",
|
| "call_id": item.call_id,
|
| "output": json.dumps(result),
|
| }
|
| )
|
| elif item.type == "message":
|
| print(f" message: {response.output_text!r}")
|
| else:
|
| print(f" {item.type} item")
|
| if not outputs:
|
| print("\nSmoke test OK: the model answered after the tool calls.")
|
| return 0
|
| print(f" -> {len(outputs)} function call(s) in this turn; sending results")
|
| next_input = outputs
|
| previous_id = response.id
|
| print("Stopped: no final answer within the call limit.", file=sys.stderr)
|
| return 1
|
|
|
|
|
| if __name__ == "__main__":
|
| raise SystemExit(main())
|
|
|
