Skip to content


本篇文章中包含如何扩展 torch.nn, torch.autograd和 使用我们的 C 库编写自定义的C扩展。

扩展 torch.autograd

如果你想要添加一个新的 Operationautograd的话,你的Operation需要继承 class Functionautograd使用Function计算结果和梯度,同时编码 operation的历史。每个新的 operation(function) 都需要实现三个方法:

  • __init__ (optional) - 如果你的operation包含非Variable参数,那么就将其作为__init__的参数传入到operation中。例如:AddConstant Function加一个常数,Transpose Function需要指定哪两个维度需要交换。如果你的operation不需要额外的参数,你可以忽略__init__

  • forward() - 在里面写执行此operation的代码。可以有任意数量的参数。如果你对某些参数指定了默认值,则这些参数是可传可不传的。记住:forward()的参数只能是Variable。函数的返回值既可以是 Variable也可以是Variablestuple。同时,请参考 Function[function]的 doc,查阅有哪些 方法是只能在forward中调用的。

  • backward() - 梯度计算公式。 参数的个数和forward返回值的个数一样,每个参数代表传回到此operation的梯度. backward()的返回值的个数应该和此operation输入的个数一样,每个返回值对应了输入值的梯度。如果operation的输入不需要梯度,或者不可导,你可以返回None。 如果forward()存在可选参数,你可以返回比输入更多的梯度,只是返回的是None

下面是 Linear 的实现代码:

# Inherit from Function
class Linear(Function):

    # bias is an optional argument
    def forward(self, input, weight, bias=None):
        self.save_for_backward(input, weight, bias)
        output =
        if bias is not None:
            output += bias.unsqueeze(0).expand_as(output)
        return output

    # This function has only a single output, so it gets only one gradient
    def backward(self, grad_output):
        # This is a pattern that is very convenient - at the top of backward
        # unpack saved_tensors and initialize all gradients w.r.t. inputs to
        # None. Thanks to the fact that additional trailing Nones are
        # ignored, the return statement is simple even when the function has
        # optional inputs.
        input, weight, bias = self.saved_tensors
        grad_input = grad_weight = grad_bias = None

        # These needs_input_grad checks are optional and there only to
        # improve efficiency. If you want to make your code simpler, you can
        # skip them. Returning gradients for inputs that don't require it is
        # not an error.
        if self.needs_input_grad[0]:
            grad_input =
        if self.needs_input_grad[1]:
            grad_weight = grad_output.t().mm(input)
        if bias is not None and self.needs_input_grad[2]:
            grad_bias = grad_output.sum(0).squeeze(0)

        return grad_input, grad_weight, grad_bias

现在,为了可以更简单的使用自定义的operation,我们建议将其用一个简单的 helper function 包装起来。 functions:

def linear(input, weight, bias=None):
    # First braces create a Function object. Any arguments given here
    # will be passed to __init__. Second braces will invoke the __call__
    # operator, that will then use forward() to compute the result and
    # return it.
    return Linear()(input, weight, bias)

你可能想知道你刚刚实现的 backward方法是否正确的计算了梯度。你可以使用 小的有限的差分进行数值估计。

from torch.autograd import gradcheck

# gradchek takes a tuple of tensor as input, check if your gradient
# evaluated with these tensors are close enough to numerical
# approximations and returns True if they all verify this condition.
input = (Variable(torch.randn(20,20).double(), requires_grad=True),)
test = gradcheck.gradcheck(Linear(), input, eps=1e-6, atol=1e-4)

扩展 torch.nn

nn 包含两种接口 - modules和他们的functional版本。通过这两个接口,你都可以扩展nn。但是我们建议,在扩展layer的时候,使用modules, 因为modules保存着参数和buffer。如果不需要参数的话,那么建议使用functional(激活函数,pooling,这些都不需要参数)。


增加一个模块(module)。 由于nn重度使用autograd。所以,添加一个新module需要实现一个 用来执行 计算 和 计算梯度 的Function。从现在开始,假定我们想要实现一个Linear module,记得之前我们已经实现了一个Linear Funciton。 只需要很少的代码就可以完成这个工作。 现在,我们需要实现两个方法:

  • __init__ (optional) - 输入参数,例如kernel sizes, numbers of features, 等等。同时初始化 parametersbuffers

  • forward() - 实例化一个执行operationFunction,使用它执行operation。和functional wrapper(上面实现的那个简单的wrapper)十分类似。

Linear module实现代码:

class Linear(nn.Module):
    def __init__(self, input_features, output_features, bias=True):
        self.input_features = input_features
        self.output_features = output_features

        # nn.Parameter is a special kind of Variable, that will get
        # automatically registered as Module's parameter once it's assigned
        # as an attribute. Parameters and buffers need to be registered, or
        # they won't appear in .parameters() (doesn't apply to buffers), and
        # won't be converted when e.g. .cuda() is called. You can use
        # .register_buffer() to register buffers.
        # nn.Parameters can never be volatile and, different than Variables,
        # they require gradients by default.
        self.weight = nn.Parameter(torch.Tensor(input_features, output_features))
        if bias:
            self.bias = nn.Parameter(torch.Tensor(output_features))
            # You should always register all possible parameters, but the
            # optional ones can be None if you want.
            self.register_parameter('bias', None)

        # Not a very smart way to initialize weights, 0.1)
        if bias is not None:
  , 0.1)

    def forward(self, input):
        # See the autograd section for explanation of what happens here.
        return Linear()(input, self.weight, self.bias)


Coming soon. For now you can find an example at GitHub.