Tensor

utils/tensor

Helper module for Tensor processing.

These functions and classes are only used internally, meaning an end-user shouldn’t need to access anything here.

• utils/tensor

  • static

    • .Tensor

      • new Tensor(...args)

      • .Symbol.iterator() ⇒ Iterator

      • ._getitem(index) ⇒ Tensor

      • .indexOf(item) ⇒ number

      • ._subarray(index, iterSize, iterDims) ⇒ Tensor

      • .item() ⇒ number

      • .tolist() ⇒ Array

      • .sigmoid() ⇒ Tensor

      • .sigmoid_() ⇒ Tensor

      • .transpose(...dims) ⇒ Tensor

      • .sum([dim], keepdim) ⇒

      • .norm([p], [dim], [keepdim]) ⇒ Tensor

      • .normalize_([p], [dim]) ⇒ Tensor

      • .normalize([p], [dim]) ⇒ Tensor

      • .stride() ⇒ Array.<number>

      • .squeeze([dim]) ⇒

      • .squeeze_()

      • .unsqueeze(dim) ⇒

      • .unsqueeze_()

      • .flatten_()

      • .flatten(start_dim, end_dim) ⇒

      • .view(...dims) ⇒ Tensor

    • .transpose(tensor, axes) ⇒ Tensor

    • .interpolate(input, size, mode, align_corners) ⇒ Tensor

    • .mean_pooling(last_hidden_state, attention_mask) ⇒ Tensor

    • .cat(tensors, dim) ⇒ Tensor

    • .stack(tensors, dim) ⇒ Tensor

    • .std_mean(input, dim, correction, keepdim) ⇒ Array.<Tensor>

    • .mean(input, dim, keepdim) ⇒

    • .dynamicTimeWarping(matrix) ⇒ Array.<Array<number>>

    • .ones(size)

    • .ones_like(tensor) ⇒

  • inner

    • ~ONNXTensor : Object

    • ~reshape(data, dimensions) ⇒ *

      • ~reshapedArray : any

    • ~DataArray : *

    • ~NestArray : *

---

utils/tensor.Tensor

Kind: static class of utils/tensor

• .Tensor

  • new Tensor(...args)

  • .Symbol.iterator() ⇒ Iterator

  • ._getitem(index) ⇒ Tensor

  • .indexOf(item) ⇒ number

  • ._subarray(index, iterSize, iterDims) ⇒ Tensor

  • .item() ⇒ number

  • .tolist() ⇒ Array

  • .sigmoid() ⇒ Tensor

  • .sigmoid_() ⇒ Tensor

  • .transpose(...dims) ⇒ Tensor

  • .sum([dim], keepdim) ⇒

  • .norm([p], [dim], [keepdim]) ⇒ Tensor

  • .normalize_([p], [dim]) ⇒ Tensor

  • .normalize([p], [dim]) ⇒ Tensor

  • .stride() ⇒ Array.<number>

  • .squeeze([dim]) ⇒

  • .squeeze_()

  • .unsqueeze(dim) ⇒

  • .unsqueeze_()

  • .flatten_()

  • .flatten(start_dim, end_dim) ⇒

  • .view(...dims) ⇒ Tensor

---

new Tensor(...args)

Create a new Tensor or copy an existing Tensor.

Param	Type
...args	*

---

tensor.Symbol.iterator() ⇒ <code> Iterator </code>

Returns an iterator object for iterating over the tensor data in row-major order. If the tensor has more than one dimension, the iterator will yield subarrays.

Kind: instance method of Tensor Returns: Iterator - An iterator object for iterating over the tensor data in row-major order.

---

tensor._getitem(index) ⇒ <code> Tensor </code>

Index into a Tensor object.

Kind: instance method of Tensor Returns: Tensor - The data at the specified index.

Param	Type	Description
index	number	The index to access.

---

tensor.indexOf(item) ⇒ <code> number </code>

Kind: instance method of Tensor Returns: number - The index of the first occurrence of item in the tensor data.

Param	Type	Description
item	number | bigint	The item to search for in the tensor

---

tensor._subarray(index, iterSize, iterDims) ⇒ <code> Tensor </code>

Kind: instance method of Tensor

Param	Type
index	number
iterSize	number
iterDims	any

---

tensor.item() ⇒ <code> number </code>

Returns the value of this tensor as a standard JavaScript Number. This only works for tensors with one element. For other cases, see Tensor.tolist().

Kind: instance method of Tensor Returns: number - The value of this tensor as a standard JavaScript Number. Throws:

• Error If the tensor has more than one element.

---

tensor.tolist() ⇒ <code> Array </code>

Convert tensor data to a n-dimensional JS list

Kind: instance method of Tensor

---

tensor.sigmoid() ⇒ <code> Tensor </code>

Return a new Tensor with the sigmoid function applied to each element.

Kind: instance method of Tensor Returns: Tensor - The tensor with the sigmoid function applied.

---

tensor.sigmoid_() ⇒ <code> Tensor </code>

Applies the sigmoid function to the tensor in place.

Kind: instance method of Tensor Returns: Tensor - Returns this.

---

tensor.transpose(...dims) ⇒ <code> Tensor </code>

Return a transposed version of this Tensor, according to the provided dimensions.

Kind: instance method of Tensor Returns: Tensor - The transposed tensor.

Param	Type	Description
...dims	number	Dimensions to transpose.

---

tensor.sum([dim], keepdim) ⇒

Returns the sum of each row of the input tensor in the given dimension dim.

Kind: instance method of Tensor Returns: The summed tensor

Param	Type	Default	Description
[dim]	number		The dimension or dimensions to reduce. If null, all dimensions are reduced.
keepdim	boolean	false	Whether the output tensor has dim retained or not.

---

tensor.norm([p], [dim], [keepdim]) ⇒ <code> Tensor </code>

Returns the matrix norm or vector norm of a given tensor.

Kind: instance method of Tensor Returns: Tensor - The norm of the tensor.

Param	Type	Default	Description
[p]	number | string	'fro'	The order of norm
[dim]	number		Specifies which dimension of the tensor to calculate the norm across. If dim is None, the norm will be calculated across all dimensions of input.
[keepdim]	boolean	false	Whether the output tensors have dim retained or not.

---

tensor.normalize_([p], [dim]) ⇒ <code> Tensor </code>

Performs L_p normalization of inputs over specified dimension. Operates in place.

Kind: instance method of Tensor Returns: Tensor - this for operation chaining.

Param	Type	Default	Description
[p]	number	2	The exponent value in the norm formulation
[dim]	number	1	The dimension to reduce

---

tensor.normalize([p], [dim]) ⇒ <code> Tensor </code>

Performs L_p normalization of inputs over specified dimension.

Kind: instance method of Tensor Returns: Tensor - The normalized tensor.

Param	Type	Default	Description
[p]	number	2	The exponent value in the norm formulation
[dim]	number	1	The dimension to reduce

---

tensor.stride() ⇒ <code> Array. < number > </code>

Compute and return the stride of this tensor. Stride is the jump necessary to go from one element to the next one in the specified dimension dim.

Kind: instance method of Tensor Returns: Array.<number> - The stride of this tensor.

---

tensor.squeeze([dim]) ⇒

Returns a tensor with all specified dimensions of input of size 1 removed.

NOTE: The returned tensor shares the storage with the input tensor, so changing the contents of one will change the contents of the other. If you would like a copy, use tensor.clone() before squeezing.

Kind: instance method of Tensor Returns: The squeezed tensor

Param	Type	Default	Description
[dim]	number		If given, the input will be squeezed only in the specified dimensions.

---

tensor.squeeze_()

In-place version of @see Tensor.squeeze

Kind: instance method of Tensor

---

tensor.unsqueeze(dim) ⇒

Returns a new tensor with a dimension of size one inserted at the specified position.

NOTE: The returned tensor shares the same underlying data with this tensor.

Kind: instance method of Tensor Returns: The unsqueezed tensor

Param	Type	Default	Description
dim	number		The index at which to insert the singleton dimension

---

tensor.unsqueeze_()

In-place version of @see Tensor.unsqueeze

Kind: instance method of Tensor

---

tensor.flatten_()

In-place version of @see Tensor.flatten

Kind: instance method of Tensor

---

tensor.flatten(start_dim, end_dim) ⇒

Flattens input by reshaping it into a one-dimensional tensor. If start_dim or end_dim are passed, only dimensions starting with start_dim and ending with end_dim are flattened. The order of elements in input is unchanged.

Kind: instance method of Tensor Returns: The flattened tensor.

Param	Type	Default	Description
start_dim	number	0	the first dim to flatten
end_dim	number		the last dim to flatten

---

tensor.view(...dims) ⇒ <code> Tensor </code>

Returns a new tensor with the same data as the self tensor but of a different shape.

Kind: instance method of Tensor Returns: Tensor - The tensor with the same data but different shape

Param	Type	Description
...dims	number	the desired size

---

utils/tensor.transpose(tensor, axes) ⇒ <code> Tensor </code>

Transposes a tensor according to the provided axes.

Kind: static method of utils/tensor Returns: Tensor - The transposed tensor.

Param	Type	Description
tensor	any	The input tensor to transpose.
axes	Array	The axes to transpose the tensor along.

---

utils/tensor.interpolate(input, size, mode, align_corners) ⇒ <code> Tensor </code>

Interpolates an Tensor to the given size.

Kind: static method of utils/tensor Returns: Tensor - The interpolated tensor.

Param	Type	Description
input	Tensor	The input tensor to interpolate. Data must be channel-first (i.e., [c, h, w])
size	Array.<number>	The output size of the image
mode	string	The interpolation mode
align_corners	boolean	Whether to align corners.

---

utils/tensor.mean_pooling(last_hidden_state, attention_mask) ⇒ <code> Tensor </code>

Perform mean pooling of the last hidden state followed by a normalization step.

Kind: static method of utils/tensor Returns: Tensor - Returns a new Tensor of shape [batchSize, embedDim].

Param	Type	Description
last_hidden_state	Tensor	Tensor of shape [batchSize, seqLength, embedDim]
attention_mask	Tensor	Tensor of shape [batchSize, seqLength]

---

utils/tensor.cat(tensors, dim) ⇒ <code> Tensor </code>

Concatenates an array of tensors along a specified dimension.

Kind: static method of utils/tensor Returns: Tensor - The concatenated tensor.

Param	Type	Description
tensors	Array.<Tensor>	The array of tensors to concatenate.
dim	number	The dimension to concatenate along.

---

utils/tensor.stack(tensors, dim) ⇒ <code> Tensor </code>

Stack an array of tensors along a specified dimension.

Kind: static method of utils/tensor Returns: Tensor - The stacked tensor.

Param	Type	Description
tensors	Array.<Tensor>	The array of tensors to stack.
dim	number	The dimension to stack along.

---

utils/tensor.std_mean(input, dim, correction, keepdim) ⇒ <code> Array. < Tensor > </code>

Calculates the standard deviation and mean over the dimensions specified by dim. dim can be a single dimension or null to reduce over all dimensions.

Kind: static method of utils/tensor Returns: Array.<Tensor> - A tuple of (std, mean) tensors.

Param	Type	Description
input	Tensor	the input tenso
dim	number | null	the dimension to reduce. If None, all dimensions are reduced.
correction	number	difference between the sample size and sample degrees of freedom. Defaults to Bessel's correction, correction=1.
keepdim	boolean	whether the output tensor has dim retained or not.

---

utils/tensor.mean(input, dim, keepdim) ⇒

Returns the mean value of each row of the input tensor in the given dimension dim.

Kind: static method of utils/tensor Returns: A new tensor with means taken along the specified dimension.

Param	Type	Description
input	Tensor	the input tensor.
dim	number | null	the dimension to reduce.
keepdim	boolean	whether the output tensor has dim retained or not.

---

utils/tensor.dynamicTimeWarping(matrix) ⇒ <code> Array. < Array < number > > </code>

Measures similarity between two temporal sequences (e.g., input audio and output tokens to generate token-level timestamps).

Kind: static method of utils/tensor

Param	Type
matrix	Tensor

---

utils/tensor.ones(size)

Returns a tensor filled with the scalar value 1, with the shape defined by the variable argument size.

Kind: static method of utils/tensor

Param	Type	Description
size	Array.<number>	A sequence of integers defining the shape of the output tensor.

---

utils/tensor.ones_like(tensor) ⇒

Returns a tensor filled with the scalar value 1, with the same size as input.

Kind: static method of utils/tensor Returns: The ones tensor.

Param	Type	Description
tensor	Tensor	The size of input will determine size of the output tensor.

---

utils/tensor~ONNXTensor : <code> Object </code>

Kind: inner constant of utils/tensor

---

utils/tensor~reshape(data, dimensions) ⇒ <code> * </code>

Reshapes a 1-dimensional array into an n-dimensional array, according to the provided dimensions.

Kind: inner method of utils/tensor Returns: * - The reshaped array.

Param	Type	Description
data	Array.<T>	The input array to reshape.
dimensions	DIM	The target shape/dimensions.

Example

Copied

reshape([10                    ], [1      ]); // Type: number[]      Value: [10]
  reshape([1, 2, 3, 4            ], [2, 2   ]); // Type: number[][]    Value: [[1, 2], [3, 4]]
  reshape([1, 2, 3, 4, 5, 6, 7, 8], [2, 2, 2]); // Type: number[][][]  Value: [[[1, 2], [3, 4]], [[5, 6], [7, 8]]]
  reshape([1, 2, 3, 4, 5, 6, 7, 8], [4, 2   ]); // Type: number[][]    Value: [[1, 2], [3, 4], [5, 6], [7, 8]]

---

reshape~reshapedArray : <code> any </code>

Kind: inner property of reshape

---

utils/tensor~DataArray : <code> * </code>

Kind: inner typedef of utils/tensor

---

utils/tensor~NestArray : <code> * </code>

This creates a nested array of a given type and depth (see examples).

Kind: inner typedef of utils/tensor Example

Copied

NestArray<string, 1>; // string[]

Example

Copied

NestArray<number, 2>; // number[][]

Example

Copied

NestArray<string, 3>; // string[][][] etc.