When the skipped-tool guard could not run the planned call (not a literal call, or arguments the tool does not take) it delivered the answer it had held back, which was given without the tool and could be a made-up value. The planned call is now recorded as a failed tool result and the model answers from that, so the user hears that it could not check or do it. M0 kit: a case passes only when the final answer uses the tool result (13 degrees, not a made-up 24) or, for the no-tool request, says it cannot do it; the no-tool request now counts. The kit mirrors the engine's failed-result path, and a failing model moves on to the next candidate on these stricter checks. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
220 lines
11 KiB
Python
220 lines
11 KiB
Python
"""Re-measure tool calling with the production fix on the real models.
|
|
|
|
The first M0 run caught both models answering past a tool they had been
|
|
given (a made-up "24도, 맑음" instead of get_weather, "설정했습니다" without
|
|
set_timer). Jarvis now fixes that in two layers, and this test runs both
|
|
layers' real code against the model under test:
|
|
|
|
1. Prompt: ``TOOL_HONESTY_RULES`` and ``tool_honesty_reminder`` from
|
|
``jarvis/system_prompt.py`` are added to the voice prompt.
|
|
2. Skipped-tool guard: Jarvis's planner (``jarvis/reply/planner.py``, its
|
|
real prompt) plans the turn on the same model. When the model answers
|
|
without a tool and the plan names one as a concrete call, the engine runs
|
|
that call itself and the model answers from the result. Here the tool
|
|
result is canned, as in the first run's round trip.
|
|
|
|
The Jarvis files come from ``jarvis_src/`` next to ``m0lib`` in the packed
|
|
kit, or from the repository's ``src/jarvis`` when run from a checkout.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import importlib.util
|
|
import json
|
|
import sys
|
|
import types
|
|
from pathlib import Path
|
|
from typing import Any
|
|
|
|
from m0lib.common import StageResult, say
|
|
from m0lib.llm_bench import NO_THINK, TOOLS, VOICE_SYSTEM
|
|
from m0lib.llm_client import ChatClient
|
|
|
|
_KIT = Path(__file__).resolve().parent.parent
|
|
_PKG = "m0_jarvis"
|
|
|
|
# (prompt, expected tool or None, argument check, canned tool result, check of the final answer)
|
|
# The final check is what makes a case pass: the right tool call is not enough
|
|
# if the spoken answer then ignores the result or claims something not done.
|
|
REFUSAL = "refusal"
|
|
FIX_CASES: list[tuple[str, str | None, Any, dict | None, str | None]] = [
|
|
("서울 날씨 어때?", "get_weather", lambda a: "서울" in str(a.get("city", "")) or "Seoul" in str(a.get("city", "")),
|
|
{"city": "서울", "temp_c": 13, "sky": "흐리고 비", "rain_mm": 2}, "13"),
|
|
("부산 지금 비 와?", "get_weather", lambda a: "부산" in str(a.get("city", "")) or "Busan" in str(a.get("city", "")),
|
|
{"city": "부산", "temp_c": 17, "sky": "맑음", "rain_mm": 0}, "17"),
|
|
("나 지금 서울인데 우산 챙겨야 할까?", "get_weather",
|
|
lambda a: "서울" in str(a.get("city", "")) or "Seoul" in str(a.get("city", "")),
|
|
{"city": "서울", "temp_c": 13, "sky": "흐리고 비", "rain_mm": 2}, "비"),
|
|
("5분 타이머 맞춰 줘.", "set_timer", lambda a: int(a.get("minutes", 0)) == 5, {"ok": True, "minutes": 5}, "5"),
|
|
("10분 뒤에 알려 줘.", "set_timer", lambda a: int(a.get("minutes", 0)) == 10, {"ok": True, "minutes": 10}, "10"),
|
|
("두 번째 모니터에 뭐 떠 있어?", "view_screen", lambda a: int(a.get("monitor", 0)) == 2,
|
|
{"monitor": 2, "seen": "유튜브 재생 화면, 제목 '오늘의 뉴스'"}, "뉴스"),
|
|
("안녕, 오늘 기분 어때?", None, None, None, None),
|
|
# No tool can do this: passing means saying so, not claiming it was done.
|
|
("거실 불 꺼 줘.", None, None, None, REFUSAL),
|
|
]
|
|
|
|
# Words a Korean answer uses to say it cannot do something. The kit's prompts
|
|
# are fixed Korean sentences, so a word check is enough here (the product
|
|
# itself stays language-agnostic).
|
|
_CANNOT = ("수 없", "수가 없", "못 ", "못해", "못합", "못했", "어렵", "지원하지", "기능이 없", "기능은 없", "불가")
|
|
|
|
|
|
def case_for(prompt: str) -> tuple:
|
|
return next(c for c in FIX_CASES if c[0] == prompt)
|
|
|
|
|
|
def is_refusal(text: str) -> bool:
|
|
return any(w in text for w in _CANNOT)
|
|
|
|
|
|
def jarvis_src() -> Path | None:
|
|
for p in (_KIT / "jarvis_src", _KIT.parent.parent / "src" / "jarvis"):
|
|
if (p / "system_prompt.py").is_file() and (p / "reply" / "planner.py").is_file():
|
|
return p
|
|
return None
|
|
|
|
|
|
def load_jarvis(src: Path, client: ChatClient) -> tuple[types.ModuleType, types.ModuleType]:
|
|
"""Import Jarvis's real system_prompt and planner modules from ``src``.
|
|
|
|
The planner's two dependencies are replaced: ``jarvis.debug`` (reads the
|
|
whole Jarvis config) by a no-op logger, and ``jarvis.llm.call_llm_direct``
|
|
(Ollama's API) by a call to the OpenAI-compatible server under test.
|
|
"""
|
|
def call_llm_direct(base_url, chat_model, system_prompt, user_content, timeout_sec=10.0,
|
|
thinking=False, num_ctx=0, temperature=None):
|
|
a = client.chat([{"role": "system", "content": system_prompt}, {"role": "user", "content": user_content}],
|
|
max_tokens=300, extra=NO_THINK)
|
|
return a["text"]
|
|
|
|
# A private package name, so a real ``jarvis`` package in the same process
|
|
# (the repository's tests) is left alone.
|
|
root = _PKG
|
|
for name, path in ((root, src), (f"{root}.reply", src / "reply")):
|
|
pkg = types.ModuleType(name)
|
|
pkg.__path__ = [str(path)]
|
|
sys.modules[name] = pkg
|
|
debug = types.ModuleType(f"{root}.debug")
|
|
debug.debug_log = lambda *a, **k: None
|
|
llm = types.ModuleType(f"{root}.llm")
|
|
llm.call_llm_direct = call_llm_direct
|
|
llm.OLLAMA_NUM_CTX = 8192
|
|
sys.modules[f"{root}.debug"], sys.modules[f"{root}.llm"] = debug, llm
|
|
|
|
loaded = []
|
|
for name, path in ((f"{root}.system_prompt", src / "system_prompt.py"),
|
|
(f"{root}.reply.planner", src / "reply" / "planner.py")):
|
|
spec = importlib.util.spec_from_file_location(name, path)
|
|
mod = importlib.util.module_from_spec(spec)
|
|
sys.modules[name] = mod
|
|
spec.loader.exec_module(mod)
|
|
loaded.append(mod)
|
|
return loaded[0], loaded[1]
|
|
|
|
|
|
def _call_ok(call: dict, want: str, check) -> bool:
|
|
if call.get("name") != want:
|
|
return False
|
|
try:
|
|
return bool(check(json.loads(call.get("arguments") or "{}")))
|
|
except Exception:
|
|
return False
|
|
|
|
|
|
def run_case(c: ChatClient, system: str, planner, case: tuple) -> dict:
|
|
"""One try through prompt layer -> planner -> guard -> final answer, as the engine does it."""
|
|
prompt, want, check, result, final_check = case
|
|
tool_names = [t["function"]["name"] for t in TOOLS]
|
|
cfg = types.SimpleNamespace(planner_enabled=True, ollama_base_url="m0", planner_model=c.model,
|
|
ollama_chat_model=c.model, planner_timeout_sec=60.0)
|
|
row: dict[str, Any] = {"prompt": prompt, "expected": want}
|
|
plan = planner.plan_query(cfg, prompt, "", planner.planner_tool_catalog(TOOLS))
|
|
row["plan"] = plan
|
|
msgs = [{"role": "system", "content": system}, {"role": "user", "content": prompt}]
|
|
a = c.chat(msgs, max_tokens=200, tools=TOOLS, extra=NO_THINK)
|
|
calls = a["tool_calls"]
|
|
row.update(model_calls=calls, model_text=a["text"][:200], first_tool=a["first_tool"], ttft=a["ttft"])
|
|
# Text next to a tool call is read aloud before the result in streaming voice.
|
|
row["text_with_call"] = bool(calls and a["text"].strip())
|
|
row["model_ok"] = (not calls) if want is None else bool(calls) and _call_ok(calls[0], want, check)
|
|
|
|
call, tool_content = (calls[0], None) if calls else (None, None)
|
|
row["guard_ran"] = False
|
|
steps = [] if calls else planner.named_tool_steps(plan, tool_names)
|
|
if steps:
|
|
# Skipped-tool guard, as in the engine: run the literal planned call, or
|
|
# if it cannot run, answer from a failed tool result (never the guess).
|
|
parsed = planner.parse_concrete_tool_step(steps[0], TOOLS)
|
|
if parsed:
|
|
call = {"name": parsed[0], "arguments": json.dumps(parsed[1], ensure_ascii=False)}
|
|
row["guard_ran"] = True
|
|
row["guard_call"] = call
|
|
else:
|
|
call = {"name": steps[0].split()[0], "arguments": "{}"}
|
|
tool_content = (f"Error: the planned call `{steps[0]}` could not be run, so nothing was fetched or "
|
|
"done. Tell the user plainly that you could not check or do it right now; do not "
|
|
"guess a value or claim it was done.")
|
|
row["guard_failed_call"] = steps[0]
|
|
|
|
final = a["text"]
|
|
if call is not None:
|
|
if tool_content is None:
|
|
if result is not None and want is not None and _call_ok(call, want, check):
|
|
tool_content = json.dumps(result, ensure_ascii=False)
|
|
else:
|
|
tool_content = f"Error: {call['name']} cannot do that request."
|
|
msgs += [{"role": "assistant", "content": "", "tool_calls": [
|
|
{"id": "call_1", "type": "function", "function": {"name": call["name"], "arguments": call["arguments"]}}]},
|
|
{"role": "tool", "tool_call_id": "call_1", "content": tool_content}]
|
|
b = c.chat(msgs, max_tokens=120, tools=TOOLS, extra=NO_THINK)
|
|
final = b["text"]
|
|
row["final_text"] = final[:200]
|
|
|
|
if want is not None:
|
|
row["final_uses_result"] = bool(final_check) and final_check in final
|
|
row["ok"] = bool(call) and _call_ok(call, want, check) and row["final_uses_result"]
|
|
elif final_check == REFUSAL:
|
|
row["ok"] = not calls and is_refusal(final)
|
|
else:
|
|
row["ok"] = not calls
|
|
return row
|
|
|
|
|
|
def tools_passed(summary: dict | None) -> bool:
|
|
"""Every case passed: the right tool call (by the model or the guard) and a final
|
|
answer that uses the tool's result, or an honest refusal where no tool can help."""
|
|
return bool(summary) and summary["total"] > 0 and summary["ok"] == summary["total"]
|
|
|
|
|
|
def test_tools_fixed(r: StageResult, c: ChatClient, tries: int = 3, key: str = "tools_fixed") -> dict | None:
|
|
say(" - 도구 호출 (수정 적용: 지시문 + 계획 + 건너뛴 도구 실행)")
|
|
src = jarvis_src()
|
|
if src is None:
|
|
r.error(key, "jarvis_src 폴더가 없습니다")
|
|
return None
|
|
sp, planner = load_jarvis(src, c)
|
|
names = [t["function"]["name"] for t in TOOLS]
|
|
system = f"{VOICE_SYSTEM}\n\n{sp.TOOL_HONESTY_RULES}\n\n{sp.tool_honesty_reminder(names)}"
|
|
rows = []
|
|
for case in FIX_CASES:
|
|
prompt, want = case[0], case[1]
|
|
for i in range(tries):
|
|
try:
|
|
row = run_case(c, system, planner, case)
|
|
except Exception as e:
|
|
row = {"prompt": prompt, "expected": want, "ok": False, "error": str(e)[:300]}
|
|
row["try"] = i + 1
|
|
rows.append(row)
|
|
r.save()
|
|
scored = rows # every case counts, the no-tool refusal too
|
|
summary = {"model_ok": sum(1 for x in scored if x.get("model_ok")),
|
|
"ok": sum(1 for x in scored if x.get("ok")),
|
|
"guard_ran": sum(1 for x in rows if x.get("guard_ran")),
|
|
"text_with_call": sum(1 for x in rows if x.get("text_with_call")),
|
|
"total": len(scored)}
|
|
r[key] = {"cases": rows, "summary": summary, "system": system}
|
|
say(f" 모델 스스로 {summary['model_ok']}/{summary['total']}, 계획 실행 포함 {summary['ok']}/{summary['total']}, "
|
|
f"도구 호출과 같이 쓴 답 {summary['text_with_call']}번")
|
|
return summary
|