Neural Network Acceleration Basics

Neural Network Acceleration Basics

Implementation process

instruction execution time

processor

Computational bottleneck calculation code

import torch
import torchvision
from tqdm import tqdm

DEVICE = "cuda:0"

model = torchvision.models.mobilenet_v2(pretrained=True)
model = model.to(DEVICE)

with torch.no_grad():
    data = torch.rand(size=[1,3,224,224])
    for i in tqdm(range(1024)):
        o = model.forward(data.to(DEVICE))

    data = torch.rand(size=[128,3,224,224])

    for i in tqdm(range(128)):
        o = model.forward(data.to(DEVICE))