模型计算性能测试代码（计算模型FPS、FLOPs和参数量）

卖报的大地主

已于 2024-04-15 16:55:01 修改

阅读量1k

点赞数 7

分类专栏：深度学习文章标签：深度学习 pytorch 人工智能

于 2024-03-13 22:42:20 首次发布

本文链接：https://blog.csdn.net/qq_43456016/article/details/136693851

版权

深度学习专栏收录该内容

54 篇文章

订阅专栏

本文介绍了如何使用Python库如numpy和tqdm在PyTorch中对深度学习模型进行性能测量，包括计算FPS（每秒帧数）、测量FLOPs（浮点运算次数）和Params（参数数量），以评估模型效率。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

import numpy as np
from tqdm import tqdm
from thop import profile

# 加载模型
args = parse_args()
model = init_model(args.config, args.checkpoint, device=args.device)
device = torch.device("cuda:0")
model.to(device)
model.eval()

# 计算FPS
dummy_input = torch.randn(1, 3, 256, 256, dtype=torch.float32).to(device)
starter, ender = torch.cuda.Event(enable_timing=True), torch.cuda.Event(enable_timing=True)
repetitions = 300
timings=np.zeros((repetitions,1))
#GPU-WARM-UP
for _ in tqdm(range(10)):
    _ = model(dummy_input)
# MEASURE PERFORMANCE
with torch.no_grad():
    for rep in tqdm(range(repetitions)):
        starter.record()
        # _= inference_model(model, dummy_input)
        _ = model(dummy_input)
        ender.record()
        # WAIT FOR GPU SYNC
        torch.cuda.synchronize()
        curr_time = starter.elapsed_time(ender)
        timings[rep] = curr_time
mean_syn = np.sum(timings) / repetitions
std_syn = np.std(timings)
mean_fps = 1000. / mean_syn
print(' * Mean@1 {mean_syn:.3f}ms Std@5 {std_syn:.3f}ms FPS@1 {mean_fps:.2f}'.format(mean_syn=mean_syn, std_syn=std_syn, mean_fps=mean_fps))
print(mean_syn)

# 计算FLOPs和Params 
flops, params = profile(model, inputs=(dummy_input,))
print('FLOPs = ' + str(flops/1000**3) + 'G')
print('Params = ' + str(params/1000**2) + 'M')