diff --git a/api/mind2web_benchmark.json b/api/mind2web_benchmark.json new file mode 100644 index 00000000..7e429b60 --- /dev/null +++ b/api/mind2web_benchmark.json @@ -0,0 +1,52 @@ +[ + { + "index": 3, + "original_task_id": "mind2web_3_1776818853", + "api_task_id": "80a5c0420dc244a8afccb3b89742089d", + "task_description": "Show me the coming soon AMC Artisan Films in amctheatres", + "reference_length": 6, + "status": "succeeded", + "queue_time_sec": 0.02, + "execution_time_sec": 40.26, + "total_time_sec": 40.26, + "result": { + "task_id": "80a5c0420dc244a8afccb3b89742089d", + "status": "succeeded", + "success": true, + "execution_time": 39.63884091377258, + "result": "**Coming Soon AMC Artisan Films**\n\nHere are all the upcoming AMC Artisan Films currently listed:\n\n| Film Title | Release Date |\n|------------|--------------|\n| **Hamlet** | April 10, 2026 |\n| **The AI Doc: Or How I Became an Apocaloptimist** | April 17, 2026 |\n| **Lorne** | April 17, 2026 |\n| **Mother Mary** | April 17, 2026 |\n| **MICHAEL** | April 24, 2026 |\n\nAll films are available for ticket purchase through AMC Theatres website.", + "error": null, + "raw_response": { + "success": true, + "result": "**Coming Soon AMC Artisan Films**\n\nHere are all the upcoming AMC Artisan Films currently listed:\n\n| Film Title | Release Date |\n|------------|--------------|\n| **Hamlet** | April 10, 2026 |\n| **The AI Doc: Or How I Became an Apocaloptimist** | April 17, 2026 |\n| **Lorne** | April 17, 2026 |\n| **Mother Mary** | April 17, 2026 |\n| **MICHAEL** | April 24, 2026 |\n\nAll films are available for ticket purchase through AMC Theatres website.", + "browser_view": "https://face-veteran-investigate-daniel.trycloudflare.com" + } + }, + "timestamp": "2026-04-22T03:48:13.806637" + }, + { + "index": 4, + "original_task_id": "mind2web_4_1776818893", + "api_task_id": "6dffc48b440d4a56b82d13f2a3a60b3e", + "task_description": "Go to the page with help in choosing a solar energy product for homes in tesla", + "reference_length": 5, + "status": "succeeded", + "queue_time_sec": 0.01, + "execution_time_sec": 251.22, + "total_time_sec": 251.22, + "result": { + "task_id": "6dffc48b440d4a56b82d13f2a3a60b3e", + "status": "succeeded", + "success": true, + "execution_time": 249.99302887916565, + "result": "**Tesla Solar Energy Products Guide - Research Summary**\n\nDue to persistent access restrictions (403 errors) on Tesla's official website preventing direct navigation to product pages (/solarpanels, /powerwall, /why-solar), I gathered comprehensive information from third-party sources.\n\n**Tesla Home Solar Products Available:**\n\n**1. Solar Panels (Traditional)**\n- **Cost:** $8,100-$17,000 after federal tax credit (for average home)\n- **Efficiency:** ~2.74-$3.30 per watt; typical systems achieve $36k before incentives\n- **Best For:** Budget-conscious homeowners prioritizing performance over aesthetics\n\n**2. Solar Roof**\n- **Cost:** $54,000-$72,000 for average 9kW system ($6-8/watt)\n- **Efficiency:** Only 8-14% (vs traditional panels at 20-22%) - rated poorly at 3/25\n- **Warranty:** 25 years equipment & efficiency coverage\n- **Degradation:** Poor - only guarantees 95% after year 5, drops to ~85% after year 25\n- **Durability:** Class F wind (110 mph), Class 4 hail (2-inch), Class A fire rating\n- **Best For:** Homeowners prioritizing aesthetics over performance; willing to pay premium for integrated roof appearance\n\n**3. Powerwall (Energy Storage)**\n- **Cost:** $13,配 k-$16配 k each\n- **Best For:** Energy independence; backup power; maximizing solar self-consumption\n\n**Key Recommendations:**\n✅ Choose Traditional Solar Panels if:\n - You want better value (~half the cost of Solar Roof)\n - Performance/efficiency is your priority\n - You already have a functional roof\n \n❌ Avoid Solar Roof unless:\n - Aesthetics are your #1 priority (looks like regular shingles)\n - You need a new roof anyway and want integrated solution\n - Budget is not a concern ($6k+ premium over traditional panels)\n \n⚠️ Important Notes:\n - Solar Roof has poor efficiency degradation compared to competitors like Maxeon (which offers better warranties up to 40 years)\n - Labor warranty comes from installer (~10 years industry average) not Tesla directly\n - Snow slides off easily despite no published snow load rating\n \nTo proceed with ordering or getting exact quotes for your home location:\n→ Visit tesla.com directly or contact local installers through third-party platforms like EnergySage or SunPower dealers.", + "error": null, + "raw_response": { + "success": true, + "result": "**Tesla Solar Energy Products Guide - Research Summary**\n\nDue to persistent access restrictions (403 errors) on Tesla's official website preventing direct navigation to product pages (/solarpanels, /powerwall, /why-solar), I gathered comprehensive information from third-party sources.\n\n**Tesla Home Solar Products Available:**\n\n**1. Solar Panels (Traditional)**\n- **Cost:** $8,100-$17,000 after federal tax credit (for average home)\n- **Efficiency:** ~2.74-$3.30 per watt; typical systems achieve $36k before incentives\n- **Best For:** Budget-conscious homeowners prioritizing performance over aesthetics\n\n**2. Solar Roof**\n- **Cost:** $54,000-$72,000 for average 9kW system ($6-8/watt)\n- **Efficiency:** Only 8-14% (vs traditional panels at 20-22%) - rated poorly at 3/25\n- **Warranty:** 25 years equipment & efficiency coverage\n- **Degradation:** Poor - only guarantees 95% after year 5, drops to ~85% after year 25\n- **Durability:** Class F wind (110 mph), Class 4 hail (2-inch), Class A fire rating\n- **Best For:** Homeowners prioritizing aesthetics over performance; willing to pay premium for integrated roof appearance\n\n**3. Powerwall (Energy Storage)**\n- **Cost:** $13,配 k-$16配 k each\n- **Best For:** Energy independence; backup power; maximizing solar self-consumption\n\n**Key Recommendations:**\n✅ Choose Traditional Solar Panels if:\n - You want better value (~half the cost of Solar Roof)\n - Performance/efficiency is your priority\n - You already have a functional roof\n \n❌ Avoid Solar Roof unless:\n - Aesthetics are your #1 priority (looks like regular shingles)\n - You need a new roof anyway and want integrated solution\n - Budget is not a concern ($6k+ premium over traditional panels)\n \n⚠️ Important Notes:\n - Solar Roof has poor efficiency degradation compared to competitors like Maxeon (which offers better warranties up to 40 years)\n - Labor warranty comes from installer (~10 years industry average) not Tesla directly\n - Snow slides off easily despite no published snow load rating\n \nTo proceed with ordering or getting exact quotes for your home location:\n→ Visit tesla.com directly or contact local installers through third-party platforms like EnergySage or SunPower dealers.", + "browser_view": "https://face-veteran-investigate-daniel.trycloudflare.com" + } + }, + "timestamp": "2026-04-22T03:52:25.037604" + } +] \ No newline at end of file diff --git a/api/test-api.py b/api/test-api.py new file mode 100644 index 00000000..1eef5b68 --- /dev/null +++ b/api/test-api.py @@ -0,0 +1,197 @@ +import requests +import time +import json +from datasets import load_dataset +from datetime import datetime + +# Конфигурация API +API_URL = "http://localhost:8088/api/browser/tasks" +HEADERS = {"Content-Type": "application/json"} + +# Загружаем датасет +dataset = load_dataset("iMeanAI/Mind2Web-Live", split="train") + +# Для теста берем первые N задач (замените на полный датасет при необходимости) +TEST_SIZE = 10 # или len(dataset) для полного бенчмарка +dataset = dataset.select(range(TEST_SIZE)) + +print(f"Загружено задач: {len(dataset)}") +print(f"Поля: {dataset[0].keys()}\n") +cnt = 3 +results = [] + +for idx, item in enumerate(dataset): + if cnt > 0: + cnt -=1 + continue + # Поля из датасета + task_desc = item['task'] # Описание задачи + ref_length = item['reference_task_length'] # Эталонная длина в шагах + evaluation = item['evaluation'] # Критерии оценки + + # ID задачи (используем index + timestamp для уникальности) + task_id_orig = f"mind2web_{idx}_{int(time.time())}" + + print(f"\n[{idx + 1}/{len(dataset)}] Task: {task_desc[:70]}...") + print(f" Эталонная длина: {ref_length} шагов") + + start_time = time.time() + + # 1. Создаем задачу через API + try: + resp = requests.post( + API_URL, + json={ + "task": task_desc, + "timeout": 300, # Увеличим таймаут для сложных задач + "metadata": { + "source": "mind2web", + "reference_length": ref_length + } + }, + headers=HEADERS, + timeout=10 + ) + + if resp.status_code != 202: + print(f" ❌ Ошибка создания задачи: {resp.status_code}") + print(f" Ответ: {resp.text}") + continue + + api_task_id = resp.json()["task_id"] + created_at = time.time() + queue_time = created_at - start_time + + print(f" 📝 Task ID: {api_task_id} | Очередь: {queue_time:.2f}с") + + # 2. Ожидание завершения с прогрессом + status = "queued" + poll_count = 0 + while status in ["queued", "running"]: + time.sleep(2) # Интервал опроса + poll_count += 1 + + try: + status_resp = requests.get(f"{API_URL}/{api_task_id}", timeout=5) + if status_resp.status_code == 200: + status_data = status_resp.json() + status = status_data.get("status", "unknown") + + # Показываем прогресс каждые 5 опросов + if poll_count % 5 == 0: + elapsed = time.time() - start_time + print(f" ⏳ Статус: {status} | Прошло: {elapsed:.1f}с") + except Exception as e: + print(f" ⚠️ Ошибка опроса: {e}") + pass + + end_time = time.time() + execution_time = end_time - start_time + + # 3. Получение результата + result_resp = requests.get(f"{API_URL}/{api_task_id}/result", timeout=10) + + result_data = None + if result_resp.status_code == 200: + try: + result_data = result_resp.json() + except: + result_data = result_resp.text + + # 4. Запись метрик + result = { + "index": idx, + "original_task_id": task_id_orig, + "api_task_id": api_task_id, + "task_description": task_desc, + "reference_length": ref_length, + "status": status, + "queue_time_sec": round(queue_time, 2), + "execution_time_sec": round(execution_time, 2), + "total_time_sec": round(end_time - start_time, 2), + "result": result_data, + "timestamp": datetime.now().isoformat() + } + results.append(result) + timestamp = datetime.now().strftime("%Y%m%d_%H%M%S") + filename = f"mind2web_benchmark.json" + with open(filename, "w", encoding="utf-8") as f: + json.dump(results, f, indent=2, ensure_ascii=False) + # Эмодзи статуса + status_emoji = "✅" if status == "succeeded" else "❌" + print(f" {status_emoji} Статус: {status} | Время: {execution_time:.1f}с") + + except requests.exceptions.Timeout: + print(f" ❌ Таймаут при создании задачи") + except Exception as e: + print(f" ❌ Ошибка: {type(e).__name__}: {e}") + continue + +# Сохранение детальных результатов +timestamp = datetime.now().strftime("%Y%m%d_%H%M%S") +filename = f"mind2web_benchmark_{timestamp}.json" + +with open(filename, "w", encoding="utf-8") as f: + json.dump(results, f, indent=2, ensure_ascii=False) + +print("\n" + "=" * 60) +print("📊 ИТОГОВЫЕ МЕТРИКИ СКОРОСТИ") +print("=" * 60) + +# Статистика по статусам +completed = [r for r in results if r["status"] == "completed"] +failed = [r for r in results if r["status"] == "failed"] +unknown = [r for r in results if r["status"] not in ["completed", "failed"]] + +print(f"\n📈 СТАТУСЫ:") +print(f" Всего задач: {len(results)}") +print(f" ✅ Успешно: {len(completed)} ({len(completed) / max(len(results), 1) * 100:.1f}%)") +print(f" ❌ Провалено: {len(failed)} ({len(failed) / max(len(results), 1) * 100:.1f}%)") +if unknown: + print(f" ❓ Неизвестный статус: {len(unknown)}") + +if completed: + total_times = [r["total_time_sec"] for r in completed] + queue_times = [r["queue_time_sec"] for r in completed] + exec_times = [r["execution_time_sec"] for r in completed] + + print(f"\n⏱️ ВРЕМЯ ВЫПОЛНЕНИЯ:") + print(f" Среднее: {sum(total_times) / len(total_times):.2f} сек") + print(f" Медиана (p50): {sorted(total_times)[len(total_times) // 2]:.2f} сек") + if len(total_times) >= 20: + print(f" p95: {sorted(total_times)[int(len(total_times) * 0.95)]:.2f} сек") + print(f" Мин: {min(total_times):.2f} сек") + print(f" Макс: {max(total_times):.2f} сек") + + print(f"\n📊 ПРОИЗВОДИТЕЛЬНОСТЬ:") + print(f" Среднее время в очереди: {sum(queue_times) / len(queue_times):.2f} сек") + tasks_per_hour = 3600 / (sum(total_times) / len(total_times)) + print(f" Скорость выполнения: {tasks_per_hour:.1f} задач/час") + + # Эффективность относительно эталонной длины + if all("reference_length" in r for r in completed): + avg_ref_length = sum(r["reference_length"] for r in completed) / len(completed) + time_per_step = (sum(total_times) / len(total_times)) / avg_ref_length + print(f" Среднее время на шаг: {time_per_step:.2f} сек") + +print(f"\n💾 Результаты сохранены в: {filename}") + +# Создание краткого отчета для сравнения +summary = { + "benchmark": "Online-Mind2Web", + "timestamp": timestamp, + "api_endpoint": API_URL, + "total_tasks": len(results), + "completed": len(completed), + "failed": len(failed), + "success_rate": len(completed) / max(len(results), 1) * 100, + "avg_time_sec": sum(total_times) / len(total_times) if completed else None, + "median_time_sec": sorted(total_times)[len(total_times) // 2] if completed else None, + "tasks_per_hour": 3600 / (sum(total_times) / len(total_times)) if completed else None +} + +summary_file = f"mind2web_summary_{timestamp}.json" +with open(summary_file, "w", encoding="utf-8") as f: + json.dump(summary, f, indent=2, ensure_ascii=False) + +print(f"📋 Краткий отчет сохранен в: {summary_file}") \ No newline at end of file