Python module
sequential
This module contains both Sequential and ModuleList containers.
For the legacy graph-based sequential container, see legacy/sequential.
A Module for a sequence of tensor transformations.
ModuleList
class max.nn.sequential.ModuleList(iterable=(), /)
A Module subclass which is locally a list container.
ModuleList instances will use the stringified integer index of their
submodules as the name of the module for the purposes of
qualified paths.
For example:
from max.nn import Linear, Sequential
model = Sequential(
Linear(5, 10),
Linear(10, 5),
)
assert dict(model.parameters).keys() == {
"0.weight", "0.bias", "1.weight", "1.bias"
}children
Iterates over the direct child modules of the Module.
-
Yields:
-
(name, module)pairs, wherenameis the attribute name of the child on the module.
Sequential
class max.nn.sequential.Sequential(*modules)
A Module subclass which holds a sequence of unary modules.
A unary Module is one whose forward() method has the signature:
def forward(self, x: Tensor) -> Tensor: ...Sequential is itself a unary Module. Its forward() method
computes the result of applying each of its child modules
in sequence to its input.
For example, this will apply a linear transformation up to a dimension of 10, apply a LayerNorm, and then apply a final linear transformation to reduce back to the input dimension of 5:
from max.tensor import Tensor
from max.nn import Linear, Sequential
model = Sequential(
Linear(5, 10),
Linear(10, 5),
)
result = model(Tensor.ones([5]))
assert result.shape == [5]-
Parameters:
-
modules (T)
forward()
forward(x)
Applies the contained modules in order.
For example, this code creates a sequence of linear transformations which each increase the dimension of the input by 5.
The input tensor must have dim 5. The intermediate applications will result in intermediate tensors of dim 10 and 15 respectively, and the final result will have dim 20:
from max.tensor import Tensor
from max.nn import Linear, Sequential
hidden_dims = [5, 10, 15, 20]
model = Sequential(*(
Linear(in_dim, out_dim) for in_dim, out_dim in
zip(hidden_dims, hidden_dims[1:])
))
result = model(Tensor.ones([5]))
assert result.shape == [20]Was this page helpful?
Thank you! We'll create more content like this.
Thank you for helping us improve!