PyTorch torch.nn.Parameter Function

PyTorch torch.nn 参考手册PyTorch torch.nn Reference Manual

torch.nn.ParameterIt is a learnable parameter tensor in PyTorch.

It is a wrapper that converts an ordinary tensor into a learnable parameter and is automatically added to the module's parameter list.

Function Definition

torch.nn.Parameter(data=None, requires_grad=True)

Usage Examples

Example 1: Basic Usage

Example

import torch
import torch.nn as nn

class MyModule(nn.Module):
    def __init__(self):
        super(MyModule, self).__init__()
        # Create a learnable parameter
        self.weight = nn.Parameter(torch.randn(10, 5))
        self.bias = nn.Parameter(torch.zeros(5))

    def forward(self, x):
        return x @ self.weight.t() + self.bias

model = MyModule()
print("Parameters:", list(model.named_parameters()))
print("Weight shape:", model.weight.shape)

Example 2: Alternative to register_parameter

Example

import torch
import torch.nn as nn

# Method 1: Use nn.Parameter (recommended)
class Net1(nn.Module):
    def __init__(self):
        super().__init__()
        self.param = nn.Parameter(torch.ones(5))

# Method 2: Use register_parameter (equivalent)
class Net2(nn.Module):
    def __init__(self):
        super().__init__()
        self.register_parameter('param', nn.Parameter(torch.ones(5)))

print("Method 1 parameters:", list(Net1().named_parameters()))
print("Method 2 parameters:", list(Net2().named_parameters()))

Example 3: Non-learnable Parameters

Example

import torch
import torch.nn as nn

# requires_grad=False creates a non-learnable parameter
param_no_grad = nn.Parameter(torch.ones(5), requires_grad=False)

print("Trainable:", param_no_grad.requires_grad)
print("Still a Parameter:", type(param_no_grad))

Use Cases

  • Custom layers: Implement learnable weights
  • Special modules: Non-standard parameters
  • Direct access: Convenient operation

Tip: nn.Parameter is a subclass of tensor and is automatically added to parameters().

PyTorch torch.nn 参考手册PyTorch torch.nn Reference Manual

Other Extensions