（beta）torch_npu.utils.save_async

接口原型

torch_npu.utils.save_async(obj, f, pickle_module=pickle, pickle_protocol=DEFAULT_PROTOCOL, _use_new_zipfile_serialization=True, _disable_byteorder_record=False, model=None)

功能描述

异步保存一个对象到一个硬盘文件上。

参数说明

支持的型号

调用示例

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
import torch
import torch.nn as nn
import os
import torch_npu

input = torch.tensor([1.,2.,3.,4.]).npu()
torch_npu.utils.save_async(input, "save_tensor.pt")
model = nn.Sequential(
    nn.Linear(100, 50),
    nn.ReLU(),
    nn.Linear(50, 20),
    nn.ReLU(),
    nn.Linear(20, 5),
    nn.ReLU()
)
model = model.npu()
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.SGD(model.parameters(), lr=0.01)
for epoch in range(3):
    for step in range(3):
        input_data = torch.ones(6400, 100).npu()
        labels = torch.randint(0, 5, (6400,)).npu()
        outputs = model(input_data)
        loss = criterion(outputs, labels)
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()
        save_path = os.path.join(f"model_{epoch}_{step}.path")
        torch_npu.utils.save_async(model, save_path, model=model)
父主题： torch_npu.utils