View Source Nx.Tensor (Nx v0.9.0)

The tensor struct and the behaviour for backends.

Nx.Tensor is a generic container for multidimensional data structures. It contains the tensor type, shape, and names. The data itself is a struct that points to a backend responsible for controlling the data. The backend behaviour is described in Nx.Backend.

The tensor has the following fields:

  • :data - the tensor backend and its data
  • :shape - the tensor shape
  • :type - the tensor type
  • :names - the tensor names
  • :vectorized_axes - a tuple that encodes names and sizes for vectorization

In general it is discouraged to access those fields directly. Use the functions in the Nx module instead. Backends have to access those fields but it cannot update them, except for the :data field itself.

Summary

Types

axes()
axis()
data()
name()
shape()
t()
t(data)
type()

Types

Link to this type

axes()

View Source 
@type axes() :: [axis()]
Link to this type

axis()

View Source 
@type axis() :: name() | integer()
Link to this type

data()

View Source 
@type data() :: Nx.Backend.t()
Link to this type

name()

View Source 
@type name() :: atom()
Link to this type

shape()

View Source 
@type shape() :: tuple()
Link to this type

t()

View Source 
@type t() :: %Nx.Tensor{
  data: data(),
  names: [name()],
  shape: shape(),
  type: type(),
  vectorized_axes: term()
}
Link to this type

t(data)

View Source 
@type t(data) :: %Nx.Tensor{
  data: data,
  names: [name()],
  shape: shape(),
  type: type(),
  vectorized_axes: term()
}
Link to this type

type()

View Source 
@type type() :: Nx.Type.t()