pytorch加载自定义网络权重的实现

时间:2022-09-09 10:28:37

在将自定义的网络权重加载到网络中时,报错:

AttributeError: 'dict' object has no attribute 'seek'. You can only torch.load from a file that is seekable. Please pre-load the data into a buffer like io.BytesIO and try to load from it instead.

我们一步一步分析。

模型网络权重保存额代码是:torch.save(net.state_dict(),'net.pkl')

(1)查看获取模型权重的源码:

pytorch源码:net.state_dict()

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
def state_dict(self, destination=None, prefix='', keep_vars=False):
  r"""Returns a dictionary containing a whole state of the module.
 
  Both parameters and persistent buffers (e.g. running averages) are
  included. Keys are corresponding parameter and buffer names.
 
  Returns:
    dict:
      a dictionary containing a whole state of the module
 
  Example::
 
    >>> module.state_dict().keys()
    ['bias', 'weight']
 
  """

将网络中所有的状态保存到一个字典中了,我自己构建的就是一个字典,没问题!

(2)查看保存模型权重的源码:

pytorch源码:torch.save()

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
def save(obj, f, pickle_module=pickle, pickle_protocol=DEFAULT_PROTOCOL):
  """Saves an object to a disk file.
 
  See also: :ref:`recommend-saving-models`
 
  Args:
    obj: saved object
    f: a file-like object (has to implement write and flush) or a string
      containing a file name
    pickle_module: module used for pickling metadata and objects
    pickle_protocol: can be specified to override the default protocol
 
  .. warning::
    If you are using Python 2, torch.save does NOT support StringIO.StringIO
    as a valid file-like object. This is because the write method should return
    the number of bytes written; StringIO.write() does not do this.
 
    Please use something like io.BytesIO instead.

函数功能是将字典保存为磁盘文件(二进制数据),那么我们在torch.load()时,就是在内存中加载二进制数据,这就是报错点。

解决方案:将字典保存为BytesIO文件之后,模型再net.load_state_dict()

?
1
2
3
#b为自定义的字典
torch.save(b,'new.pkl')
net.load_state_dict(torch.load(b))

解决方法很简单,主要记录解决思路。

以上这篇pytorch加载自定义网络权重的实现就是小编分享给大家的全部内容了,希望能给大家一个参考,也希望大家多多支持服务器之家。

原文链接:https://blog.csdn.net/qq_34789262/article/details/83376374