| """Run the new plain-English witness prompt directly against the live LLM |
| (stateless /v1/chat/completions) so we can verify the fix produces distinct |
| sensible English answers before restarting the FastAPI server.""" |
| import json, sys, time, urllib.request |
| from pathlib import Path |
| REPO = Path(__file__).resolve().parent.parent |
| sys.path.insert(0, str(REPO)) |
| from config import load_settings |
|
|
| settings = load_settings() |
| out = REPO / "runtime" / "tmp" / "witness_fix_verify.txt" |
| out.parent.mkdir(parents=True, exist_ok=True) |
|
|
| WITNESS = { |
| "name": "Harriet Moss", |
| "occupation": "passer-by", |
| "personality": {"style": "nervous", "confidence": "uncertain", "quirk": "remembers colors better than faces"}, |
| "current_summary": "A passer-by near Junction 74 saw an ordinary commuter who only loosely matched the notice.", |
| "stable_facts": ["witness was at Junction 74"], |
| } |
|
|
| SYSTEM = ( |
| "You are roleplaying a witness in an English-language detective game. " |
| "Speak only English. Reply in one or two short sentences. Use only the " |
| "facts the user gives you. Never invent details. If you don't know, " |
| "say you don't know." |
| ) |
|
|
|
|
| def build_user(question, history): |
| history_block = ( |
| "\n".join(f" Detective: {q}\n You: {a}" for q, a in history) |
| if history else " (no prior questions)" |
| ) |
| stable_block = ", ".join(WITNESS["stable_facts"]) |
| personality_block = ", ".join(f"{k}: {v}" for k, v in WITNESS["personality"].items()) |
| return ( |
| f"You are {WITNESS['name']}, a {WITNESS['occupation']} ({personality_block}).\n" |
| f"What you saw / know: {WITNESS['current_summary']}\n" |
| f"Stable facts: {stable_block}\n" |
| f"Conversation so far:\n{history_block}\n" |
| f"The detective now asks: {question!r}\n" |
| f"Reply in character, in English, in one or two short sentences." |
| ) |
|
|
|
|
| def call(messages): |
| payload = { |
| "model": settings.minicpm_quantization or settings.llm_model, |
| "messages": messages, |
| "temperature": 0.55, |
| "max_tokens": 120, |
| "stream": False, |
| } |
| req = urllib.request.Request( |
| "http://127.0.0.1:19060/v1/chat/completions", |
| data=json.dumps(payload).encode("utf-8"), |
| headers={"Content-Type": "application/json"}, |
| method="POST", |
| ) |
| start = time.perf_counter() |
| with urllib.request.urlopen(req, timeout=90) as resp: |
| data = json.loads(resp.read().decode("utf-8")) |
| return data["choices"][0]["message"]["content"].strip(), time.perf_counter() - start |
|
|
|
|
| history = [] |
| lines = [] |
| for q in ["what did you see", "where was he", "what was he carrying", "what time was this"]: |
| user = build_user(q, history) |
| answer, elapsed = call([{"role": "system", "content": SYSTEM}, {"role": "user", "content": user}]) |
| lines.append(f"Q ({elapsed:.2f}s): {q}") |
| lines.append(f"A: {answer}") |
| lines.append("") |
| history.append((q, answer)) |
|
|
| out.write_text("\n".join(lines), encoding="utf-8") |
| print(f"wrote: {out}") |
|
|
