import urllib.request, json, time, sys try: sys.stdout.reconfigure(encoding='utf-8') except: pass BASE = "http://127.0.0.1:8000" prompt = "Write a Python function to calculate fibonacci numbers efficiently using memoization. Include type hints and docstring." payload = json.dumps({ "model": "m", "messages": [ {"role": "system", "content": "You are a helpful assistant."}, {"role": "user", "content": prompt} ], "max_tokens": 500, "temperature": 0.0 }).encode('utf-8') req = urllib.request.Request( f"{BASE}/v1/chat/completions", data=payload, headers={"Content-Type": "application/json"} ) print("Sending request...") t0 = time.time() resp = json.loads(urllib.request.urlopen(req, timeout=300).read()) dt = time.time() - t0 u = resp.get("usage", {}) tokens = u.get("completion_tokens", 0) speed = tokens / dt if dt > 0 else 0 print(f"\n=== 122B Benchmark ===") print(f"Time: {dt:.1f}s") print(f"Completion Tokens: {tokens}") print(f"Speed: {speed:.2f} t/s") print(f"\n--- Response Preview ---") print(resp["choices"][0]["message"]["content"][:300])