Delete benchmark_resnet50.py

benchmark_resnet50.py

-import torch
-import torchvision.models as models
-import time
-import json
-
-def run_resnet50_benchmark():
-    """运行 ResNet50 推理基准测试，batch_size=64"""
-    print("--- PyTorch ResNet50 Benchmark (batch_size=64) ---")
-
-    # 1. 检查设备 (CPU or GPU)
-    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-    print(f"Running on device: {device}")
-
-    # 2. 加载预训练的 ResNet50 模型
-    print("Loading ResNet50 model with pretrained weights...")
-    model = models.resnet50(weights=models.ResNet50_Weights.IMAGENET1K_V2).to(device)
-    # 设置为评估模式，关闭 Dropout 等训练层
-    model.eval()
-
-    # 3. 准备输入数据，batch_size=64
-    batch_size = 64
-    # ImageNet 标准输入尺寸: (batch_size, channels, height, width)
-    input_tensor = torch.randn(batch_size, 3, 224, 224).to(device)
-    print(f"Input tensor shape: {input_tensor.shape}")
-
-    # 4. 预热 (非常重要！避免首次运行的初始化开销影响结果)
-    print("Warming up...")
-    with torch.no_grad():  # 关闭梯度计算，节省内存和计算
-        for _ in range(10):
-            _ = model(input_tensor)
-
-    # 5. 正式进行基准测试
-    print("Running benchmark...")
-    num_runs = 100
-    start_time = time.time()
-
-    with torch.no_grad():
-        for _ in range(num_runs):
-            _ = model(input_tensor)
-
-    end_time = time.time()
-
-    # 6. 计算性能指标
-    total_time = end_time - start_time
-    total_images = num_runs * batch_size
-    
-    # 吞吐量: 每秒处理的图片数量
-    throughput = total_images / total_time
-    
-    # 平均延迟: 处理单张图片的平均时间 (毫秒)
-    avg_latency_ms = (total_time / total_images) * 1000
-
-    # 7. 整理结果
-    results = {
-        "model": "ResNet50",
-        "device": str(device),
-        "batch_size": batch_size,
-        "num_runs": num_runs,
-        "total_time_s": round(total_time, 4),
-        "throughput_imgs_per_sec": round(throughput, 2),
-        "avg_latency_ms_per_img": round(avg_latency_ms, 4)
-    }
-
-    # 8. 打印并保存结果到 JSON 文件
-    print("\n--- Benchmark Results ---")
-    print(json.dumps(results, indent=4))
-    
-    with open("results.json", "w") as f:
-        json.dump(results, f, indent=4)
-        
-    print(f"\nBenchmark finished. Results saved to results.json")
-
-
-if __name__ == "__main__":
-    run_resnet50_benchmark()
-