机器学习 - save和load训练好的模型

如果已经训练好了一个模型,你就可以save和load这模型。

For saving and loading models in PyTorch, there are three main methods you should be aware of.

PyTorch method What does it do?
torch.save Saves a serialized object to disk using Python's pickle utility. Models, tensors and various other Python objects like dictionaries can be saved using torch.save
torch.load Uses pickle's unpickling features to deserialize and load pickled Python object files (like models, tensors or dictionaries) into memory. You can also set which device to load the object to (CPU, GPU etc)
torch.nn.Module.load_state_dict Loads a model's parameter dictionary (model.state_dict()) using a saved state_dict() object

在 PyTorch 中,pickle 是一个用于序列化和反序列化Python对象的标准库模块。它可以将Python对象转换为字节流 (即序列化),并将字节流转换回Python对象 (即反序列化)。pickle模块在很多情况下都非常有用,特别是在保存和加载模型,保存训练中间状态等方面。

在深度学习中,经常需要保存训练好的模型或者训练过程中的中间结果,以便后续的使用或分析。PyTorch提高了方便的API来保存和加载模型,其中就包括了使用pickle模块进行对象的序列化和反序列化。


save model

python 复制代码
import torch
from pathlib import Path 

# 1. Create models directory
MODEL_PATH = Path("models")
MODEL_PATH.mkdir(parents = True, exist_ok = True)

# 2. Create model save path
MODEL_NAME = "trained_model.pth"
MODEL_SAVE_PATH = MODEL_PATH / MODEL_NAME

# 3. Save the model state dict 
print(f"Saving model to: {MODEL_SAVE_PATH}")
torch.save(obj = model_0.state_dict(),
			f = MODEL_SAVE_PATH)

就能看到 trained_model.pth 文件下载到所属的文件夹位置。


Load the saved PyTorch model

You can load it in using torch.nn.Module.load_state_dict(torch.load(f)) where f is the filepath of the saved model state_dict().

Why call torch.load() inside torch.nn.Module.load_state_dict()?

Because you only saved the model's state_dict() which is a dictionary of learned parameters and not the entire model, you first have to load the state_dict() with torch.load() and then pass that state_dict() to a new instance of the model (which is a subclass of nn.Module).

python 复制代码
# Instantiate a new instance of the model 
loaded_model_0 = LinearRegressionModel()

# Load the state_dict of the saved model
loaded_model_0.load_state_dict(torch.load(f=MODEL_SAVE_PATH))

# 结果如下
<All keys matched successfully>

测试 loaded model。

python 复制代码
# Put the loaded model into evaluation model 
loaded_model_0.eval() 

# 2. Use the inference mode context manager to make predictions
with torch.inference_mode():
  loaded_model_preds = loaded_model_0(X_test)

# Compare previous model predictions with loaded model predictions
print(y_preds == loaded_model_preds) 

# 结果如下
tensor([[True],
        [True],
        [True],
        [True],
        [True],
        [True],
        [True],
        [True],
        [True],
        [True]])

看到这了,点个赞呗~

相关推荐
瓶瓶持续学习中3 分钟前
物理AI与具身智能的技术架构区别:从空间建模到机器人动作执行
人工智能·机器人
羲云网络8 分钟前
GPT Image 2.6 精准编辑前,先排查参考图的访问权限与输入交接
图像处理·人工智能·aigc
snakeshe101016 分钟前
ChatGLM行业垂直领域微调实战:从需求分析到数据集加载完整流程
人工智能
智联视频超融合平台4 小时前
视频 AI 的误报之战,正在从调参数卷到改架构
人工智能·大模型·边缘计算·视频监控·安防
水如烟5 小时前
孤能子视角:智能的边界不在算法强弱,在可形式化性——从概率法到因果推理,选择池的边界一直在那里
人工智能
万里侯6 小时前
日志聚类之后别把 noise 扔掉:降噪应该改变视图,不抹掉证据
人工智能·微服务·ai·容器·k8s
澹锦汐6 小时前
纯前端秋日银杏与枫叶落叶色板智能聚类器:基于 K-Means 色彩量化与直方图自动配方提取
前端·人工智能·ai·语言模型·vue·react·k-means
71777714 小时前
国内 AI 研发效能工具怎么选:2026 年基于公开信息的选型框架与 Gitee 实践
人工智能·gitee
Csvn14 小时前
换供应商要改一遍代码?把模型接入收进一层网关(C04)
人工智能·aigc·agent