《PyTorch面试精华》

---

1 前言

---

2 深度学习之数学基础

---

3 PyTorch入门疑难点

---

4 PyTorch全局设置

---

5 PyTorch GPU分布式训练

---

6 向量的基础与核心

---

7 神经网络基础

---

8 计算机视觉基础知识

---

9 循环神经网络基础

---

10 注意力机制

---

11 PyTorch归一化

---

12 激活函数相关内容

---

13 思考题的答案

---

---

PyTorch网络模型的保存与加载总共分为4种情况，大家要牢记。首先，我们要定义一个模型，以便下文所使用。

import torch.nn as nn
import torch.nn.functional as F

class MyModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.nn1 = nn.Linear(2, 3)
        self.nn2 = nn.Linear(3, 6)

    def forward(self, x):
        x = F.relu(self.nn1(x))
        return F.relu(self.nn2(x))

model = MyModel()

1、保存整个模型

torch.save(model, "model.pth")

model = torch.load("model.pth", weights_only=False)

2、只保存模型参数

PyTorch模型将学习到的参数存储在一个名为state_dict的内部状态字典中，这些参数可以通过torch.save方法进行持久化。

在加载模型权重时，我们首先需要实例化模型类，因为该类定义了网络的结构。

# 保存模型
torch.save(model.state_dict(), "model_weights.pth")
# 加载模型
model2 = MyModel() 
model2.load_state_dict(torch.load("model_weights.pth", weights_only=True))
model2.eval()

3、保存与加载优化器参数

# 定义优化器
optimizer = optim.SGD(model.parameters(), lr=0.001, momentum=0.9)

# 保存优化器
EPOCH = 5
PATH = "model.pt"
LOSS = 0.4
torch.save({
            "epoch": EPOCH,
            "model_state_dict": net.state_dict(),
            "optimizer_state_dict": optimizer.state_dict(),
            "loss": LOSS,
            }, PATH)

# 加载优化器
model = MyModel()
optimizer = optim.SGD(model.parameters(), lr=0.001, momentum=0.9)

checkpoint = torch.load(PATH, weights_only=True)
model.load_state_dict(checkpoint["model_state_dict"])
optimizer.load_state_dict(checkpoint["optimizer_state_dict"])
epoch = checkpoint["epoch"]
loss = checkpoint["loss"]

model.eval()
# - or -
model.train()

4、保存与加载模型（跨设备）

4.1、Save on GPU, Load on CPU

PATH = "model.pt"

torch.save(model.state_dict(), PATH)

device = torch.device("cpu")
model = MyModel()
model.load_state_dict(torch.load(PATH, map_location=device, weights_only=True))

4.2、Save on CPU, Load on GPU

# Save
torch.save(net.state_dict(), PATH)

# Load
device = torch.device("cuda")
model = MyModel()
# Choose whatever GPU device number you want
model.load_state_dict(torch.load(PATH, map_location="cuda:0"))
# Make sure to call input = input.to(device) on any input tensors that you feed to the model
model.to(device)

4.3、Save on GPU, Load on GPU

torch.save(net.state_dict(), PATH)

device = torch.device("cuda")

model = MyModel()

model.load_state_dict(torch.load(PATH))

model.to(device) 

---