tvm.relay.nn

Neural network related operators.

Classes:

Constant(data[, span])

A constant expression in Relay.

Expr

alias of tvm.ir.expr.RelayExpr

Functions:

adaptive_avg_pool1d(data[, output_size, ...])

1D adaptive average pooling operator.

adaptive_avg_pool2d(data[, output_size, ...])

2D adaptive average pooling operator.

adaptive_avg_pool3d(data[, output_size, ...])

3D adaptive avg pooling operator.

adaptive_max_pool1d(data[, output_size, ...])

1D adaptive max pooling operator.

adaptive_max_pool2d(data[, output_size, ...])

2D adaptive max pooling operator.

adaptive_max_pool3d(data[, output_size, ...])

3D adaptive max pooling operator.

avg_pool1d(data[, pool_size, strides, ...])

1D average pooling operator.

avg_pool2d(data[, pool_size, strides, ...])

2D average pooling operator.

avg_pool2d_grad(out_grad, data[, pool_size, ...])

Gradient of 2D average pooling operator.

avg_pool3d(data[, pool_size, strides, ...])

3D average pooling operator.

batch_flatten(data)

BatchFlatten.

batch_matmul(tensor_a, tensor_b[, ...])

Compute batch matrix multiplication of tensor_a and tensor_b.

batch_norm(data, gamma, beta, moving_mean, ...)

Batch normalization layer (Ioffe and Szegedy, 2014).

batch_to_space_nd(data, block_shape, crops)

Reshape the batch dimension into spatial dimensions.

bias_add(data, bias[, axis])

add_bias operator.

bitpack(data[, bits, pack_axis, bit_axis, ...])

Tensor packing for bitserial operations.

bitserial_conv2d(data, weight[, strides, ...])

2D convolution using bitserial computation.

bitserial_dense(data, weight[, units, ...])

Bitserial Dense operator.

const(value[, dtype, span])

Create a constant value.

contrib_conv2d_gemm_weight_transform(...)

Weight Transformation part for 2D convolution with gemm algorithm.

contrib_conv2d_gemm_without_weight_transform(...)

2D convolution with gemm algorithm.

contrib_conv2d_nchwc(data, kernel[, ...])

Variant of 2D convolution.

contrib_conv2d_winograd_nnpack_weight_transform(...)

Weight Transformation part for 2D convolution with winograd algorithm.

contrib_conv2d_winograd_weight_transform(...)

Weight Transformation part for 2D convolution with winograd algorithm.

contrib_conv2d_winograd_without_weight_transform(...)

2D convolution with winograd algorithm.

contrib_conv3d_winograd_weight_transform(...)

Weight Transformation part for 3D convolution with winograd algorithm.

contrib_conv3d_winograd_without_weight_transform(...)

3D convolution with winograd algorithm.

contrib_dense_pack(data, weight[, ...])

Dense operator.

contrib_depthwise_conv2d_nchwc(data, kernel)

Variant of 2D depthwise convolution.

conv1d(data, weight[, strides, padding, ...])

1D convolution.

conv1d_transpose(data, weight[, strides, ...])

One dimensional transposed convolution operator.

conv2d(data, weight[, strides, padding, ...])

2D convolution.

conv2d_backward_weight(grad, data[, ...])

The gradient of conv2d with respect to weight.

conv2d_transpose(data, weight[, strides, ...])

Two dimensional transposed convolution operator.

conv3d(data, weight[, strides, padding, ...])

3D convolution.

conv3d_transpose(data, weight[, strides, ...])

3D transpose convolution.

correlation(data1, data2, kernel_size, ...)

Applies correlation to inputs.

cross_entropy(predictions, targets)

CrossEntropy without logits.

cross_entropy_with_logits(predictions, targets)

CrossEntropy with logits.

deformable_conv2d(data, offset, weight[, ...])

Deformable 2d convolution.

dense(data, weight[, units, out_dtype])

Dense operator.

depth_to_space(data, block_size[, layout, mode])

Convert channels into spatial blocks.

dilate(data, strides[, dilation_value])

Dilate data with given dilation value (0 by default).

dropout(data[, rate])

Applies the dropout operation to the input array.

dropout_raw(data[, rate])

Applies the dropout operation to the input array.

fast_softmax(data[, axis])

Computes softmax.

fifo_buffer(data, buffer, axis)

FIFO buffer to enable computation reuse in CNNs with sliding indow input

get_pad_tuple1d(padding)

Common code to get the 1 dimensional pad option :param padding: Padding size :type padding: Union[int, Tuple[int, ...]]

get_pad_tuple2d(padding)

Common code to get the pad option :param padding: Padding size :type padding: Union[int, Tuple[int, ...]]

get_pad_tuple3d(padding)

Common code to get the pad option :param padding: Padding size :type padding: Union[int, Tuple[int, ...]]

global_avg_pool1d(data[, layout, out_layout])

1D global average pooling operator.

global_avg_pool2d(data[, layout, out_layout])

2D global average pooling operator.

global_avg_pool3d(data[, layout, out_layout])

3D global average pooling operator.

global_max_pool1d(data[, layout, out_layout])

1D global maximum pooling operator.

global_max_pool2d(data[, layout, out_layout])

2D global maximum pooling operator.

global_max_pool3d(data[, layout, out_layout])

3D global maximum pooling operator.

group_norm(data, gamma, beta, num_groups[, ...])

Group normalization normalizes over group of channels for each training examples.

instance_norm(data, gamma, beta[, axis, ...])

Instance Normalization (Ulyanov and et al., 2016) Applies instance normalization to the n-dimensional input array.

l2_normalize(data, eps[, axis])

Perform L2 normalization on the input data

layer_norm(data, gamma, beta[, axis, ...])

Layer normalization (Lei Ba and et al., 2016).

leaky_relu(data[, alpha])

This operator takes data as input and does Leaky version of a Rectified Linear Unit.

log_softmax(data[, axis])

Computes log softmax.

lrn(data[, size, axis, bias, alpha, beta])

This operator takes data as input and does local response normalization.

matmul(tensor_a, tensor_b[, units, ...])

Matmul operator.

max_pool1d(data[, pool_size, strides, ...])

1D maximum pooling operator.

max_pool2d(data[, pool_size, strides, ...])

2D maximum pooling operator.

max_pool2d_grad(out_grad, data[, pool_size, ...])

Gradient of 2D maximum pooling operator.

max_pool3d(data[, pool_size, strides, ...])

3D maximum pooling operator.

mirror_pad(data, pad_width[, mode])

MirrorPadding

nll_loss(predictions, targets, weights[, ...])

Negative log likelihood loss.

pad(data, pad_width[, pad_value, pad_mode])

Padding

prelu(data, alpha[, axis])

This operator takes data as input and does Leaky version of a Rectified Linear Unit.

relu(data)

Rectified linear unit.

softmax(data[, axis])

Computes softmax.

space_to_batch_nd(data, block_shape, paddings)

Divide spatial dimensions of the data into a grid of blocks and interleave them into batch dim.

space_to_depth(data, block_size[, layout])

Convert spatial blocks into channels.

sparse_add(dense_mat, sparse_mat)

Computes the matrix addition of dense_mat and sparse_mat, where dense_mat is a dense matrix and sparse_mat is a sparse (CSR) namedtuple with fields data, indices, and indptr.

sparse_dense(dense_mat, sparse_mat[, sparse_lhs])

Computes the matrix multiplication of dense_mat and sparse_mat, where dense_mat is a dense matrix and sparse_mat is a sparse (either BSR or CSR) namedtuple with fields data, indices, and indptr.

sparse_transpose(x)

Computes the fast matrix transpose of x, where x is a sparse tensor in CSR format (represented as a namedtuple with fields data, indices, and indptr).

upsampling(data[, scale_h, scale_w, layout, ...])

Upsampling.

upsampling3d(data[, scale_d, scale_h, ...])

3D Upsampling.

class tvm.relay.nn.Constant(data, span=None)

A constant expression in Relay.

Parameters
  • data (tvm.nd.NDArray) – The data content of the constant expression.

  • span (Optional[tvm.relay.Span]) – Span that points to original source code.

tvm.relay.nn.Expr

alias of tvm.ir.expr.RelayExpr Attributes:

checked_type

Get the checked type of tvm.relay.Expr.

struct_info

Get the struct info field

tvm.relay.nn.adaptive_avg_pool1d(data, output_size=None, layout='NCW', out_layout='')

1D adaptive average pooling operator. This operator is experimental.

This operator takes data as input and does 1D average value calculation across each window represented by W.

In the default case, where the data_layout is NCW a data Tensor with shape (batch_size, in_channels, width), to produce an output Tensor with shape (batch_size, in_channels, output_width).

The pooling kernel and stride sizes are automatically chosen for desired output sizes.

For output_size:

If this argument is not provided, input height and width will be used as output width.

If a single integer is provided for output_size, the output size is (N x C x output_size) for any input (NCW).

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • output_size (tuple of int. optional) – Output height and width.

  • layout (str, optional) – Layout of the input.

  • out_layout (str, optional) – Layout of the output.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.adaptive_avg_pool2d(data, output_size=None, layout='NCHW', out_layout='')

2D adaptive average pooling operator. This operator is experimental.

This operator takes data as input and does 2D average value calculation across each window represented by WxH.

In the default case, where the data_layout is NCHW a data Tensor with shape (batch_size, in_channels, height, width), to produce an output Tensor with shape (batch_size, in_channels, output_height, output_width).

The pooling kernel and stride sizes are automatically chosen for desired output sizes.

For output_size:

If this argument is not provided, input height and width will be used as output height and width.

If a single integer is provided for output_size, the output size is (N x C x output_size x output_size) for any input (NCHW).

If a tuple of integers (height, width) are provided for output_size, the output size is (N x C x height x width) for any input (NCHW).

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • output_size (tuple of int. optional) – Output height and width.

  • layout (str, optional) – Layout of the input.

  • out_layout (str, optional) – Layout of the output.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.adaptive_avg_pool3d(data, output_size=None, layout='NCDHW', out_layout='')

3D adaptive avg pooling operator. This operator is experimental.

This operator takes data as input and does 3D avg value calculation across each window represented by DxWxH.

In the default case, where the data_layout is NCDHW a data Tensor with shape (batch_size, in_channels, depth, height, width), to produce an output Tensor with shape (batch_size, in_channels, output_depth, output_height, output_width).

The pooling kernel and stride sizes are automatically chosen for desired output sizes.

For output_size:

If this argument is not provided, input depth, height and width will be used as output depth, height and width.

If a single integer is provided for output_size, the output size is (N x C x output_size x output_size x output_size) for any input (NCDHW).

If a tuple of integers (depth, height, width) are provided for output_size, the output size is (N x C x depth x height x width) for any input (NCDHW).

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • output_size (tuple of int. optional) – Output height and width.

  • layout (str, optional) – Layout of the input.

  • out_layout (str, optional) – Layout of the output.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.adaptive_max_pool1d(data, output_size=None, layout='NCW', out_layout='')

1D adaptive max pooling operator. This operator is experimental.

This operator takes data as input and does 1D max value calculation across each window represented by W.

In the default case, where the data_layout is NCW a data Tensor with shape (batch_size, in_channels, width), to produce an output Tensor with shape (batch_size, in_channels, output_width).

The pooling kernel and stride sizes are automatically chosen for desired output sizes.

For output_size:

If this argument is not provided, input height and width will be used as output height and width.

If a single integer is provided for output_size, the output size is (N x C x output_size) for any input (NCW).

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • output_size (tuple of int. optional) – Output height and width.

  • layout (str, optional) – Layout of the input.

  • out_layout (str, optional) – Layout of the output.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.adaptive_max_pool2d(data, output_size=None, layout='NCHW', out_layout='')

2D adaptive max pooling operator. This operator is experimental.

This operator takes data as input and does 2D max value calculation across each window represented by WxH.

In the default case, where the data_layout is NCHW a data Tensor with shape (batch_size, in_channels, height, width), to produce an output Tensor with shape (batch_size, in_channels, output_height, output_width).

The pooling kernel and stride sizes are automatically chosen for desired output sizes.

For output_size:

If this argument is not provided, input height and width will be used as output height and width.

If a single integer is provided for output_size, the output size is (N x C x output_size x output_size) for any input (NCHW).

If a tuple of integers (height, width) are provided for output_size, the output size is (N x C x height x width) for any input (NCHW).

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • output_size (tuple of int. optional) – Output height and width.

  • layout (str, optional) – Layout of the input.

  • out_layout (str, optional) – Layout of the output.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.adaptive_max_pool3d(data, output_size=None, layout='NCDHW', out_layout='')

3D adaptive max pooling operator. This operator is experimental.

This operator takes data as input and does 3D max value calculation across each window represented by DxWxH.

In the default case, where the data_layout is NCDHW a data Tensor with shape (batch_size, in_channels, depth, height, width), to produce an output Tensor with shape (batch_size, in_channels, output_depth, output_height, output_width).

The pooling kernel and stride sizes are automatically chosen for desired output sizes.

For output_size:

If this argument is not provided, input depth, height and width will be used as output depth, height and width.

If a single integer is provided for output_size, the output size is (N x C x output_size x output_size x output_size) for any input (NCDHW).

If a tuple of integers (depth, height, width) are provided for output_size, the output size is (N x C x depth x height x width) for any input (NCDHW).

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • output_size (tuple of int. optional) – Output height and width.

  • layout (str, optional) – Layout of the input.

  • out_layout (str, optional) – Layout of the output.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.avg_pool1d(data, pool_size=(1,), strides=(1,), dilation=(1,), padding=(0,), layout='NCW', out_layout='', ceil_mode=False, count_include_pad=False)

1D average pooling operator.

This operator takes data as input and does 1D average value calculation with in pool_size sized window by striding defined by stride

In the default case, where the data_layout is NCW a data Tensor with shape (batch_size, channels, width), to produce an output Tensor.

The ceil_mode is used to take ceil or floor while computing out shape. count_include_pad indicates including or excluding padded input values in computation. This operator accepts data layout specification.

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • pool_size (int or tuple of int, optional) – The size of window for pooling.

  • strides (int or tuple of int, optional) – The strides of pooling.

  • dilation (int or tuple of int, optional) – The dilation of pooling.

  • padding (int or tuple of int, optional) – The padding for pooling.

  • layout (str, optional) – Layout of the input.

  • out_layout (Optional[str]) – Layout of the output

  • ceil_mode (bool, optional) – To enable or disable ceil while pooling.

  • count_include_pad (bool, optional) – To include padding to compute the average.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.avg_pool2d(data, pool_size=(1, 1), strides=(1, 1), dilation=(1, 1), padding=(0, 0), layout='NCHW', out_layout='', ceil_mode=False, count_include_pad=False)

2D average pooling operator.

This operator takes data as input and does 2D average value calculation with in pool_size sized window by striding defined by stride

In the default case, where the data_layout is NCHW a data Tensor with shape (batch_size, in_channels, height, width), to produce an output Tensor with the following rule:

with data of shape (b, c, h, w), pool_size (kh, kw)

\[\mbox{out}(b, c, y, x) = \frac{1}{kh * kw} \sum_{m=0}^{kh-1} \sum_{n=0}^{kw-1} \mbox{data}(b, c, \mbox{stride}[0] * y + m, \mbox{stride}[1] * x + n)\]

Padding is applied to data before the computation. ceil_mode is used to take ceil or floor while computing out shape. count_include_pad indicates including or excluding padded input values in computation. This operator accepts data layout specification.

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • pool_size (int or tuple of int, optional) – The size of window for pooling.

  • strides (tuple of int, optional) – The strides of pooling.

  • dilation (int or tuple of int, optional) – The dilation of pooling.

  • padding (tuple of int, optional) – The padding for pooling.

  • layout (str, optional) – Layout of the input.

  • out_layout (Optional[str]) – Layout of the output

  • ceil_mode (bool, optional) – To enable or disable ceil while pooling.

  • count_include_pad (bool, optional) – To include padding to compute the average.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.avg_pool2d_grad(out_grad, data, pool_size=(1, 1), strides=(1, 1), padding=(0, 0), layout='NCHW', out_layout='', ceil_mode=False, count_include_pad=False)

Gradient of 2D average pooling operator.

This operator takes out_grad and data as input and calculates gradient of avg_pool2d.

Parameters
  • out_grad (tvm.relay.Expr) – The output gradient

  • data (tvm.relay.Expr) – The input data to the operator.

  • pool_size (int or tuple of int, optional) – The size of window for pooling.

  • strides (tuple of int, optional) – The strides of pooling.

  • padding (tuple of int, optional) – The padding for pooling.

  • layout (str, optional) – Layout of the input.

  • out_layout (Optional[str]) – Layout of the output

  • ceil_mode (bool, optional) – To enable or disable ceil while pooling.

  • count_include_pad (bool, optional) – To include padding to compute the average.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.avg_pool3d(data, pool_size=(1, 1, 1), strides=(1, 1, 1), dilation=(1, 1, 1), padding=(0, 0, 0), layout='NCDHW', out_layout='', ceil_mode=False, count_include_pad=False)

3D average pooling operator.

This operator takes data as input and does 3D average value calculation with in pool_size sized window by striding defined by stride

In the default case, where the data_layout is NCDHW a data Tensor with shape (batch_size, channels, depth, height, width), to produce an output Tensor.

The ceil_mode is used to take ceil or floor while computing out shape. count_include_pad indicates including or excluding padded input values in computation. This operator accepts data layout specification.

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • pool_size (int or tuple of int, optional) – The size of window for pooling.

  • strides (tuple of int, optional) – The strides of pooling.

  • dilation (int or tuple of int, optional) – The dilation of pooling.

  • padding (tuple of int, optional) – The padding for pooling.

  • layout (str, optional) – Layout of the input.

  • out_layout (Optional[str]) – Layout of the output

  • ceil_mode (bool, optional) – To enable or disable ceil while pooling.

  • count_include_pad (bool, optional) – To include padding to compute the average.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.batch_flatten(data)

BatchFlatten.

This operator flattens all the dimensions except for the batch dimension. which results a 2D output.

For data with shape (d1, d2, ..., dk) batch_flatten(data) returns reshaped output of shape (d1, d2*...*dk).

Parameters

data (tvm.relay.Expr) – The input data to the operator.

Returns

result – The Flattened result.

Return type

tvm.relay.Expr

tvm.relay.nn.batch_matmul(tensor_a, tensor_b, out_dtype='', transpose_a=False, transpose_b=True)

Compute batch matrix multiplication of tensor_a and tensor_b.

Both tensor_a and tensor_b can be transposed. For legacy reason, we use NT format (transpose_a=False, transpose_b=True) by default.

\[\mbox{batch_matmul}(A, B)[i, :, :] = \mbox{matmul}(A[i, :, :], B[i, :, :])\]
Parameters
  • tensor_a (tvm.relay.Expr) – The first input.

  • tensor_b (tvm.relay.Expr) – The second input.

  • out_dtype (Optional[str]) – Specifies the output data type for mixed precision batch matmul.

  • transpose_a (Optional[bool] = False) – Whether the first tensor is in transposed format.

  • transpose_b (Optional[bool] = True) – Whether the second tensor is in transposed format.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.batch_norm(data, gamma, beta, moving_mean, moving_var, axis=1, epsilon=1e-05, center=True, scale=True)

Batch normalization layer (Ioffe and Szegedy, 2014). Normalizes the input at each batch, i.e. applies a transformation that maintains the mean activation close to 0 and the activation standard deviation close to 1.

\[\begin{split}data\_mean[i] = mean(data[:,i,:,...]) \\ data\_var[i] = var(data[:,i,:,...])\end{split}\]

Then compute the normalized output, which has the same shape as input, as following:

\[out[:,i,:,...] = \frac{data[:,i,:,...] - data\_mean[i]}{\sqrt{data\_var[i]+\epsilon}} * gamma[i] + beta[i]\]

Both mean and var returns a scalar by treating the input as a vector.

Assume the input has size k on axis 1, then both gamma and beta have shape (k,).

Besides the inputs and the outputs, this operator accepts two auxiliary states, moving_mean and moving_var, which are k-length vectors. They are global statistics for the whole dataset, which are updated by

moving_mean = moving_mean * momentum + data_mean * (1 - momentum)
moving_var = moving_var * momentum + data_var * (1 - momentum)

The parameter axis specifies which axis of the input shape denotes the ‘channel’ (separately normalized groups). The default is 1. Specifying -1 sets the channel axis to be the last item in the input shape.

Note

This operator can be optimized away for inference.

Parameters
  • data (tvm.relay.Expr) – Input to which batch_norm will be applied.

  • gamma (tvm.relay.Expr) – The gamma scale factor.

  • beta (tvm.relay.Expr) – The beta offset factor.

  • moving_mean (tvm.relay.Expr) – Running mean of input,

  • moving_var (tvm.relay.Expr) – Running variance of input.

  • axis (int, optional, default=1) – Specify along which shape axis the channel is specified.

  • epsilon (double, optional, default=1e-5) – Small float added to variance to avoid dividing by zero.

  • center (boolean, optional, default=True) – If True, add offset of beta to normalized tensor, If False, beta is ignored.

  • scale (boolean, optional, default=True) – If true, multiply by gamma. If False, gamma is not used. When the next layer is piecewise linear (also e.g. nn.relu), this can be disabled since the scaling will be done by the next layer.

Returns

result – Tuple of normed data (same shape as input), new running mean (k-length vector), and new running variance (k-length vector)

Return type

relay.Tuple([tvm.relay.Expr, tvm.relay.Expr, tvm.relay.Expr])

tvm.relay.nn.batch_to_space_nd(data, block_shape, crops)

Reshape the batch dimension into spatial dimensions.

Parameters
  • data (tvm.te.Tensor) – N-D with shape [batch, spatial_shape, remaining_shape]

  • block_shape (relay.Expr) – 1-D of size [M] where M is number of spatial dims, specifies block size for each spatial dimension.

  • crops (relay.Expr) – 2-D of shape [M, 2] where M is number of spatial dims, specifies [begin, end] crop size for each spatial dimension.

Returns

result – N-D Tensor with shape [batch / prod(block_shape), in_shape[1] * block_shape[0] - crops[0,0] - crops[0,1], …, in_shape[M] * block_shape[M-1] - crops[M-1, 0] - crops[M-1, 1], remaining_shape]

Return type

relay.Expr

tvm.relay.nn.bias_add(data, bias, axis=1)

add_bias operator.

Add 1D bias to the axis of data. This function is a special case of add which allows inference of shape of the bias from data.

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • bias (tvm.relay.Expr) – The bias to be added.

  • axis (int, optional) – The axis to add the bias.

Returns

result – The final result.

Return type

tvm.relay.Expr

tvm.relay.nn.bitpack(data, bits=1, pack_axis=1, bit_axis=2, pack_type='uint32', name='BitPack')

Tensor packing for bitserial operations.

The values along the input tensor’s pack_axis are quantized and packed together into the specified pack_type in a new bit axis.

For example, consider bitpacking with data to be a tensor with shape [1, 64, 128, 128], pack_axis=1, bit_axis=4, pack_type=uint8, and bits=2. The output in this case will be of shape [1, 8, 128, 128, 2]. The dimension of axis 1 has been reduced by a factor of 8 since each value is packed into an 8-bit uint8. Axis 4 is now two bitplanes representing the quantized value of the incoming data. The output tensor is now ready to be used in a bitserial operation.

Parameters
  • data (tvm.relay.expr) – The incoming tensor to be packed.

  • bits (int) – Number of bits that should be packed.

  • pack_axis (int) – Axis that should be decomposed and packed.

  • bit_axis (int) – New axis containing bitplane.

  • pack_type (str) – Datatype to pack bits into.

  • name (str, optional) – Name of the operation.

Returns

result – The packed tensor.

Return type

tvm.relay.Expr

tvm.relay.nn.bitserial_conv2d(data, weight, strides=(1, 1), padding=(0, 0), channels=None, kernel_size=(3, 3), activation_bits=1, weight_bits=1, data_layout='NCHW', kernel_layout='OIHW', pack_dtype='uint32', out_dtype='int16', unipolar=True)

2D convolution using bitserial computation.

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • weight (tvm.relay.Expr) – The weight expressions.

  • strides (tuple of int, optional) – The strides of convolution.

  • padding (tuple of int, optional) – The padding of convolution on both sides of inputs before convolution.

  • channels (int, optional) – Number of output channels of this convolution.

  • kernel_size (tuple of int, optional) – The spatial of the convolution kernel.

  • activation_bits (int) – Number of bits to pack for activations.

  • weight_bits (int) – Number of bits to pack for weights.

  • data_layout (str, optional) – Layout of the input.

  • kernel_layout (str, optional) – Layout of the kernel

  • pack_dtype (str, optional) – Datatype to pack bits into.

  • out_dtype (str, optional) – Specifies the output data type for mixed precision conv2d.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.bitserial_dense(data, weight, units=None, data_bits=1, weight_bits=1, pack_dtype='uint32', out_dtype='int16', unipolar=True)

Bitserial Dense operator. Applies matrix multiplication of two quantized matrices using a fast bitserial algorithm.

\[\]

Y = X * W

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • weight (tvm.relay.Expr) – The weight expressions.

  • units (int, optional) – Number of hidden units of the dense transformation.

  • data_bits (int) – Number of bits incoming tensor should be packed with.

  • weight_bits (int) – Number of bits weight tensor should be packed with.

  • pack_dtype (str, optional) – Datatype to pack individual bits into before computation.

  • out_dtype (str, optional) – Specifies the output data type for mixed precision dense.

  • unipolar (bool, optional) – Whether to use unipolar or bipolar quantization for inputs.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.const(value, dtype=None, span=None)

Create a constant value.

Parameters
  • value (Union[bool, int, float, numpy.ndarray, tvm.nd.NDArray]) – The constant value.

  • dtype (str, optional) – The data type of the resulting constant.

  • span (Optional[tvm.relay.Span]) – Span that points to original source code.

Note

When dtype is None, we use the following rule:

  • int maps to “int32”

  • float maps to “float32”

  • bool maps to “bool”

  • other using the same default rule as numpy.

tvm.relay.nn.contrib_conv2d_gemm_weight_transform(weights, tile_N, tile_K)

Weight Transformation part for 2D convolution with gemm algorithm.

We separate this as a single op to enable pre-compute for inference. Use this together with nn.contrib_conv2d_gemm_without_weight_transform

Parameters
  • weights (tvm.relay.Expr) – The weight expressions.

  • tile_N (int) – Tile size across N axis of the weight transformation for ConvGemm. (N = OC)

  • tile_K (int) – Tile size across K axis of the weight transformation for ConvGemm. (K = KW * KH * IC)

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.contrib_conv2d_gemm_without_weight_transform(data, weight, strides=(1, 1), padding=(0, 0), dilation=(1, 1), groups=1, channels=None, kernel_size=None, data_layout='NCHW', kernel_layout='OIHW', out_layout='', out_dtype='')

2D convolution with gemm algorithm.

The basic parameters are the same as the ones in vanilla conv2d. It assumes the weight is pre-transformed by nn.contrib_conv2d_gemm_weight_transform

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • weight (tvm.relay.Expr) – The weight expressions.

  • strides (tuple of int, optional) – The strides of convolution.

  • padding (tuple of int, optional) – The padding of convolution on both sides of inputs before convolution.

  • dilation (tuple of int, optional) – Specifies the dilation rate to be used for dilated convolution.

  • groups (int, optional) – Number of groups for grouped convolution.

  • channels (int, optional) – Number of output channels of this convolution.

  • kernel_size (tuple of int, optional) – The spatial of the convolution kernel.

  • data_layout (str, optional) – Layout of the input.

  • kernel_layout (str, optional) – Layout of the weight.

  • out_layout (str, optional) – Layout of the output, by default, out_layout is the same as data_layout

  • out_dtype (str, optional) – Specifies the output data type for mixed precision conv2d.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.contrib_conv2d_nchwc(data, kernel, strides=(1, 1), padding=(0, 0), dilation=(1, 1), groups=1, channels=None, kernel_size=None, data_layout='NCHW8c', kernel_layout='OIHW', out_layout='', out_dtype='')

Variant of 2D convolution.

This operator takes the weight as the convolution kernel and convolves it with data to produce an output, following a specialized NCHWc data layout.

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • kernel (tvm.relay.Expr) – The kernel expressions.

  • strides (tuple of int, optional) – The strides of convolution.

  • padding (tuple of int, optional) – The padding of convolution on both sides of inputs before convolution.

  • dilation (tuple of int, optional) – Specifies the dilation rate to be used for dilated convolution.

  • groups (int, optional) – Number of groups for grouped convolution.

  • channels (int, optional) – Number of output channels of this convolution.

  • kernel_size (tuple of int, optional) – The spatial of the convolution kernel.

  • data_layout (str, optional) – Layout of the input.

  • kernel_layout (str, optional) – Layout of the weight.

  • out_layout (str, optional) – Layout of the output, by default, out_layout is the same as data_layout

  • out_dtype (str, optional) – Specifies the output data type for mixed precision conv2d.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.contrib_conv2d_winograd_nnpack_weight_transform(weight, convolution_algorithm, out_dtype='')

Weight Transformation part for 2D convolution with winograd algorithm.

We separate this as a single op to enable pre-compute for inference. Use this together with nn.contrib_conv2d_winograd_without_weight_transform

Parameters
  • weight (tvm.relay.Expr) – The weight expressions.

  • convolution_algorithm (int) – The Tile size of winograd. E.g. 2 for F(2x2, 3x3) and 4 for F(4x4, 3x3)

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.contrib_conv2d_winograd_weight_transform(weight, tile_size)

Weight Transformation part for 2D convolution with winograd algorithm.

We separate this as a single op to enable pre-compute for inference. Use this together with nn.contrib_conv2d_winograd_without_weight_transform

Parameters
  • weight (tvm.relay.Expr) – The weight expressions.

  • tile_size (int) – The Tile size of winograd. E.g. 2 for F(2x2, 3x3) and 4 for F(4x4, 3x3)

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.contrib_conv2d_winograd_without_weight_transform(data, weight, tile_size, strides=(1, 1), padding=(0, 0), dilation=(1, 1), groups=1, channels=None, kernel_size=None, data_layout='NCHW', kernel_layout='OIHW', out_layout='', out_dtype='')

2D convolution with winograd algorithm.

The basic parameters are the same as the ones in vanilla conv2d. It assumes the weight is pre-transformed by nn.contrib_conv2d_winograd_weight_transform

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • weight (tvm.relay.Expr) – The weight expressions.

  • tile_size (int) – The Tile size of winograd. E.g. 2 for F(2x2, 3x3) and 4 for F(4x4, 3x3)

  • strides (tuple of int, optional) – The strides of convolution.

  • padding (tuple of int, optional) – The padding of convolution on both sides of inputs before convolution.

  • dilation (tuple of int, optional) – Specifies the dilation rate to be used for dilated convolution.

  • groups (int, optional) – Number of groups for grouped convolution.

  • channels (int, optional) – Number of output channels of this convolution.

  • kernel_size (tuple of int, optional) – The spatial of the convolution kernel.

  • data_layout (str, optional) – Layout of the input.

  • kernel_layout (str, optional) – Layout of the weight.

  • out_layout (str, optional) – Layout of the output, by default, out_layout is the same as data_layout

  • out_dtype (str, optional) – Specifies the output data type for mixed precision conv2d.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.contrib_conv3d_winograd_weight_transform(weight, tile_size)

Weight Transformation part for 3D convolution with winograd algorithm.

We separate this as a single op to enable pre-compute for inference. Use this together with nn.contrib_conv3d_winograd_without_weight_transform

Parameters
  • weight (tvm.relay.Expr) – The weight expressions.

  • tile_size (int) – The Tile size of winograd. E.g. 2 for F(2x2x2, 3x3x3) and 4 for F(4x4x4, 3x3x3)

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.contrib_conv3d_winograd_without_weight_transform(data, weight, tile_size, strides=(1, 1, 1), padding=(0, 0, 0), dilation=(1, 1, 1), groups=1, channels=None, kernel_size=None, data_layout='NCDHW', kernel_layout='OIDHW', out_layout='', out_dtype='')

3D convolution with winograd algorithm.

The basic parameters are the same as the ones in vanilla conv3d. It assumes the weight is pre-transformed by nn.contrib_conv3d_winograd_weight_transform

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • weight (tvm.relay.Expr) – The weight expressions.

  • tile_size (int) – The Tile size of winograd. E.g. 2 for F(2x2x2, 3x3x3) and 4 for F(4x4x4, 3x3x3)

  • strides (tuple of int, optional) – The strides of convolution.

  • padding (tuple of int, optional) – The padding of convolution on both sides of inputs before convolution.

  • dilation (tuple of int, optional) – Specifies the dilation rate to be used for dilated convolution.

  • groups (int, optional) – Number of groups for grouped convolution.

  • channels (int, optional) – Number of output channels of this convolution.

  • kernel_size (tuple of int, optional) – The spatial of the convolution kernel.

  • data_layout (str, optional) – Layout of the input.

  • kernel_layout (str, optional) – Layout of the weight.

  • out_layout (str, optional) – Layout of the output, by default, out_layout is the same as data_layout

  • out_dtype (str, optional) – Specifies the output data type for mixed precision conv2d.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.contrib_dense_pack(data, weight, weight_layout='NC', units=None, out_dtype='')

Dense operator. Applies a linear transformation with packed weight

\[\]

Y = X * W^T

Parameters
  • data (tvm.relay.Expr) – The input data to the operator, of shape (batch, units_in).

  • weight (tvm.relay.Expr) – The transformed weight expressions, 3-D matrix, of shape (units // pack_weight_tile, units_in, pack_weight_tile).

  • weight_layout (str) – The layout of weight, such as “NC” or “NC8n”.

  • units (int, optional) – Number of hidden units of the dense transformation.

  • out_dtype (str, optional) – Specifies the output data type for mixed precision dense.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.contrib_depthwise_conv2d_nchwc(data, kernel, strides=(1, 1), padding=(0, 0), dilation=(1, 1), groups=1, channels=None, kernel_size=None, data_layout='NCHW8c', kernel_layout='OIHW', out_layout='', out_dtype='')

Variant of 2D depthwise convolution.

This operator takes the weight as the depthwise convolution kernel and depthwise convolves it with data to produce an output, following a specialized NCHWc data layout.

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • kernel (tvm.relay.Expr) – The kernel expressions.

  • strides (tuple of int, optional) – The strides of convolution.

  • padding (tuple of int, optional) – The padding of convolution on both sides of inputs before convolution.

  • dilation (tuple of int, optional) – Specifies the dilation rate to be used for dilated convolution.

  • groups (int, optional) – Number of groups for grouped convolution.

  • channels (int, optional) – Number of output channels of this convolution.

  • kernel_size (tuple of int, optional) – The spatial of the convolution kernel.

  • data_layout (str, optional) – Layout of the input.

  • kernel_layout (str, optional) – Layout of the weight.

  • out_layout (str, optional) – Layout of the output, by default, out_layout is the same as data_layout

  • out_dtype (str, optional) – Specifies the output data type for mixed precision conv2d.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.conv1d(data, weight, strides=1, padding=0, dilation=1, groups=1, channels=None, kernel_size=None, data_layout='NCW', kernel_layout='OIW', out_layout='', out_dtype='')

1D convolution.

This operator takes the weight as the convolution kernel and convolves it with data to produce an output.

In the default case, where the data_layout is NCW and kernel_layout is OIW, conv1d takes in a data Tensor with shape (batch_size, in_channels, width), and a weight Tensor with shape (channels, in_channels, kernel_size) to produce an output Tensor with the following rule:

\[\mbox{out}[b, c, w] = \sum_{dw, k} \mbox{data}[b, k, \mbox{strides}[0] * w + dw] * \mbox{weight}[c, k, dw]\]

Padding and dilation are applied to data and weight respectively before the computation. This operator accepts data layout specification. Semantically, the operator will convert the layout to the canonical layout (NCW for data and OIW for weight), perform the computation, then convert to the out_layout.

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • weight (tvm.relay.Expr) – The weight expressions.

  • strides (Optional[int, Tuple[int]]) – The strides of convolution.

  • padding (Optional[int, Tuple[int]]) – The padding of convolution on both sides of the input before convolution.

  • dilation (Optional[int, Tuple[int]]) – Specifies the dilation rate to be used for dilated convolution.

  • groups (Optional[int]) – Currently unused for 1D convolution.

  • channels (Optional[int]) – Number of output channels of this convolution.

  • kernel_size (Optional[int, Tuple[int]]) – The spatial dimension of the convolution kernel.

  • data_layout (Optional[str]) – Layout of the input.

  • kernel_layout (Optional[str]) – Layout of the weight.

  • out_layout (Optional[str]) – Layout of the output, by default, out_layout is the same as data_layout

  • out_dtype (Optional[str]) – Specifies the output data type for mixed precision conv2d.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.conv1d_transpose(data, weight, strides=(1,), padding=(0,), dilation=(1,), groups=1, channels=None, kernel_size=None, data_layout='NCW', kernel_layout='IOW', out_layout='', output_padding=(0,), out_dtype='')

One dimensional transposed convolution operator.

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • weight (tvm.relay.Expr) – The weight expressions.

  • strides (Tuple[int], optional) – The strides of convolution.

  • padding (Tuple[int], optional) – The padding of convolution on both sides of inputs.

  • dilation (Tuple[int], optional) – Specifies the dilation rate to be used for dilated convolution.

  • channels (int, optional) – Number of output channels of this convolution.

  • kernel_size (tuple of int, optional) – The spatial of the convolution kernel.

  • groups (int, optional) – Number of groups for grouped convolution.

  • data_layout (str, optional) – Layout of the input.

  • kernel_layout (str, optional) – Layout of the weight.

  • out_layout (Optional[str]) – Layout of the output, by default, out_layout is the same as data_layout

  • output_padding (Tuple[int], optional) – Used to disambiguate the output shape.

  • out_dtype (str, optional) – Specifies the output data type for mixed precision conv2d.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.conv2d(data, weight, strides=(1, 1), padding=(0, 0), dilation=(1, 1), groups=1, channels=None, kernel_size=None, data_layout='NCHW', kernel_layout='OIHW', out_layout='', out_dtype='')

2D convolution.

This operator takes the weight as the convolution kernel and convolves it with data to produce an output.

In the default case, where the data_layout is NCHW and kernel_layout is OIHW, conv2d takes in a data Tensor with shape (batch_size, in_channels, height, width), and a weight Tensor with shape (channels, in_channels, kernel_size[0], kernel_size[1]) to produce an output Tensor with the following rule:

\[\mbox{out}[b, c, y, x] = \sum_{dy, dx, k} \mbox{data}[b, k, \mbox{strides}[0] * y + dy, \mbox{strides}[1] * x + dx] * \mbox{weight}[c, k, dy, dx]\]

Padding and dilation are applied to data and weight respectively before the computation. This operator accepts data layout specification. Semantically, the operator will convert the layout to the canonical layout (NCHW for data and OIHW for weight), perform the computation, then convert to the out_layout.

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • weight (tvm.relay.Expr) – The weight expressions.

  • strides (Optional[int, Tuple[int]]) – The strides of convolution.

  • padding (Optional[int, Tuple[int]]) – The padding of convolution on both sides of inputs before convolution.

  • dilation (Optional[int, Tuple[int]]) – Specifies the dilation rate to be used for dilated convolution.

  • groups (Optional[int]) – Number of groups for grouped convolution.

  • channels (Optional[int]) – Number of output channels of this convolution.

  • kernel_size (Optional[int, Tuple[int]]) – The spatial of the convolution kernel.

  • data_layout (Optional[str]) – Layout of the input.

  • kernel_layout (Optional[str]) – Layout of the weight.

  • out_layout (Optional[str]) – Layout of the output, by default, out_layout is the same as data_layout

  • out_dtype (Optional[str]) – Specifies the output data type for mixed precision conv2d.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.conv2d_backward_weight(grad, data, strides=(1, 1), padding=(0, 0), dilation=(1, 1), groups=1, channels=None, kernel_size=None, grad_layout='NCHW', data_layout='NCHW', kernel_layout='OIHW', out_dtype='')

The gradient of conv2d with respect to weight.

This operator takes the output gradient grad and convolves it with data as the convolution kernel, to produce the gradient with respect to weight.

Note that the parameter kernel_size is the spatial size of the corresponding forward convolution kernel, not that of data. grad_layout and kernel_layout are the layouts of grad and the weight gradient respectively.

Other parameters are the same as the conv2d op. See its documentation for more details.

tvm.relay.nn.conv2d_transpose(data, weight, strides=(1, 1), padding=(0, 0), dilation=(1, 1), groups=1, channels=None, kernel_size=None, data_layout='NCHW', kernel_layout='IOHW', out_layout='', output_padding=(0, 0), out_dtype='')

Two dimensional transposed convolution operator.

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • weight (tvm.relay.Expr) – The weight expressions.

  • strides (Tuple[int], optional) – The strides of convolution.

  • padding (Tuple[int], optional) – The padding of convolution on both sides of inputs.

  • dilation (Tuple[int], optional) – Specifies the dilation rate to be used for dilated convolution.

  • channels (int, optional) – Number of output channels of this convolution.

  • kernel_size (tuple of int, optional) – The spatial of the convolution kernel.

  • groups (int, optional) – Number of groups for grouped convolution.

  • data_layout (str, optional) – Layout of the input.

  • kernel_layout (str, optional) – Layout of the weight.

  • out_layout (Optional[str]) – Layout of the output, by default, out_layout is the same as data_layout

  • output_padding (Tuple[int], optional) – Used to disambiguate the output shape.

  • out_dtype (str, optional) – Specifies the output data type for mixed precision conv2d.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.conv3d(data, weight, strides=(1, 1, 1), padding=(0, 0, 0), dilation=(1, 1, 1), groups=1, channels=None, kernel_size=None, data_layout='NCDHW', kernel_layout='OIDHW', out_layout='', out_dtype='')

3D convolution.

This operator takes the weight as the convolution kernel and convolves it with data to produce an output.

In the default case, where the data_layout is NCDHW and kernel_layout is OIDHW, conv3d takes in a data Tensor with shape (batch_size, in_channels, depth, height, width), and a weight Tensor with shape (channels, in_channels, kernel_size[0], kernel_size[1], kernel_size[2]) to produce an output Tensor with the following rule:

\[\mbox{out}[b, c, z, y, x] = \sum_{dz, dy, dx, k} \mbox{data}[b, k, \mbox{strides}[0] * z + dz, \mbox{strides}[1] * y + dy, \mbox{strides}[2] * x + dx] * \mbox{weight}[c, k, dz, dy, dx]\]

Padding and dilation are applied to data and weight respectively before the computation. This operator accepts data layout specification. Semantically, the operator will convert the layout to the canonical layout (NCDHW for data and OIDHW for weight), perform the computation, then convert to the out_layout.

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • weight (tvm.relay.Expr) – The weight expressions.

  • strides (Optional[Tuple[int]]) – The strides of convolution.

  • padding (Optional[int, Tuple[int]]) – The padding of convolution on both sides of inputs before convolution.

  • dilation (Optional[int, Tuple[int]]) – Specifies the dilation rate to be used for dilated convolution.

  • groups (Optional[int]) – Number of groups for grouped convolution.

  • channels (Optional[int]) – Number of output channels of this convolution.

  • kernel_size (Optional[int, Tuple[int]]) – The spatial of the convolution kernel.

  • data_layout (Optional[str]) – Layout of the input.

  • kernel_layout (Optional[str]) – Layout of the weight.

  • out_layout (Optional[str]) – Layout of the output, by default, out_layout is the same as data_layout

  • out_dtype (Optional[str]) – Specifies the output data type for mixed precision conv2d.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.conv3d_transpose(data, weight, strides=(1, 1, 1), padding=(0, 0, 0), dilation=(1, 1, 1), groups=1, channels=None, kernel_size=None, data_layout='NCDHW', kernel_layout='IODHW', out_layout='', output_padding=(0, 0, 0), out_dtype='')

3D transpose convolution.

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • weight (tvm.relay.Expr) – The weight expressions.

  • strides (Optional[Tuple[int]]) – The strides of convolution.

  • padding (Optional[int, Tuple[int]]) – The padding of convolution on both sides of inputs before convolution.

  • dilation (Optional[int, Tuple[int]]) – Specifies the dilation rate to be used for dilated convolution.

  • groups (Optional[int]) – Number of groups for grouped convolution.

  • channels (Optional[int]) – Number of output channels of this convolution.

  • kernel_size (Optional[int, Tuple[int]]) – The spatial of the convolution kernel.

  • data_layout (Optional[str]) – Layout of the input.

  • kernel_layout (Optional[str]) – Layout of the weight.

  • out_layout (Optional[str]) – Layout of the output, by default, out_layout is the same as data_layout

  • out_dtype (Optional[str]) – Specifies the output data type for mixed precision conv3d.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.correlation(data1, data2, kernel_size, max_displacement, stride1, stride2, padding, is_multiply, layout)

Applies correlation to inputs.

The correlation layer performs multiplicative patch comparisons between two feature maps. Given two multi-channel feature maps \(f_{1}, f_{2}\), with \(w\), \(h\), and \(c\) being their width, height, and number of channels, the correlation layer lets the network compare each patch from \(f_{1}\) with each patch from \(f_{2}\).

For now we consider only a single comparison of two patches. The ‘correlation’ of two patches centered at \(x_{1}\) in the first map and \(x_{2}\) in the second map is then defined as:

\[c(x_{1}, x_{2}) = \sum_{o \in [-k,k] \times [-k,k]} <f_{1}(x_{1} + o), f_{2}(x_{2} + o)>\]

for a square patch of size \(K:=2k+1\).

Note that the equation above is identical to one step of a convolution in neural networks, but instead of convolving data with a filter, it convolves data with other data. For this reason, it has no training weights.

Computing \(c(x_{1}, x_{2})\) involves \(c * K^{2}\) multiplications. Comparing all patch combinations involves \(w^{2}*h^{2}\) such computations.

Given a maximum displacement \(d\), for each location \(x_{1}\) it computes correlations \(c(x_{1}, x_{2})\) only in a neighborhood of size \(D:=2d+1\), by limiting the range of \(x_{2}\). We use strides \(s_{1}, s_{2}\), to quantize \(x_{1}\) globally and to quantize \(x_{2}\) within the neighborhood centered around \(x_{1}\).

The final output is defined by the following expression:

\[out[n, q, i, j] = c(x_{i, j}, x_{q})\]

where \(i\) and \(j\) enumerate spatial locations in \(f_{1}\), and \(q\) denotes the \(q^{th}\) neighborhood of \(x_{i,j}\).

Parameters
  • data1 (tvm.te.Tensor) – 4-D with shape [batch, channel, height, width]

  • data2 (tvm.te.Tensor) – 4-D with shape [batch, channel, height, width]

  • kernel_size (int) – Kernel size for correlation, must be an odd number

  • max_displacement (int) – Max displacement of Correlation

  • stride1 (int) – Stride for data1

  • stride2 (int) – Stride for data2 within the neightborhood centered around data1

  • padding (int or a list/tuple of 2 or 4 ints) – Padding size, or [pad_height, pad_width] for 2 ints, or [pad_top, pad_left, pad_bottom, pad_right] for 4 ints

  • is_multiply (bool) – operation type is either multiplication or substraction

  • layout (str) – layout of data1, data2 and the output

Returns

Output – 4-D with shape [batch, out_channel, out_height, out_width]

Return type

tvm.te.Tensor

tvm.relay.nn.cross_entropy(predictions, targets)

CrossEntropy without logits.

Parameters
  • predictions (tvm.relay.Expr) – The predictions.

  • targets (tvm.relay.Expr) – The targets.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.cross_entropy_with_logits(predictions, targets)

CrossEntropy with logits.

Parameters
  • predictions (tvm.relay.Expr) – The predictions.

  • targets (tvm.relay.Expr) – The targets.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.deformable_conv2d(data, offset, weight, strides=(1, 1), padding=(0, 0), dilation=(1, 1), deformable_groups=1, groups=1, channels=None, kernel_size=None, data_layout='NCHW', kernel_layout='OIHW', out_layout='', out_dtype='')

Deformable 2d convolution.

The deformable convolution operation is described in https://arxiv.org/abs/1703.06211

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • offset (tvm.relay.Expr) – The offset expressions.

  • weight (tvm.relay.Expr) – The weight expressions.

  • strides (tuple of int, optional) – The strides of convolution.

  • padding (tuple of int, optional) – The padding of convolution on both sides of inputs before convolution.

  • dilation (tuple of int, optional) – Specifies the dilation rate to be used for dilated convolution.

  • deformable_groups (int, optional) – Number of deformable groups.

  • groups (int, optional) – Number of groups for grouped convolution.

  • channels (int, optional) – Number of output channels of this convolution.

  • kernel_size (tuple of int, optional) – The spatial of the convolution kernel.

  • data_layout (str, optional) – Layout of the input.

  • kernel_layout (str, optional) – Layout of the weight.

  • out_layout (str, optional) – Layout of the output, by default, out_layout is the same as data_layout

  • out_dtype (str, optional) – Specifies the output data type for mixed precision conv2d.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.dense(data, weight, units=None, out_dtype='')

Dense operator. Applies a linear transformation

\[\]

Y = X * W^T

Parameters
  • data (tvm.relay.Expr) – The input data to the operator, of shape (d_1, d_2, …, d_n, units_in).

  • weight (tvm.relay.Expr) – The weight expressions, 2-D matrix, of shape (units, units_in).

  • units (int, optional) – Number of hidden units of the dense transformation.

  • out_dtype (str, optional) – Specifies the output data type for mixed precision dense, of shape (d_1, d_2, …, d_n, units).

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.depth_to_space(data, block_size, layout='NCHW', mode='DCR')

Convert channels into spatial blocks.

Parameters
  • data (tvm.relay.Expr) – Input data with channels divisible by block_size**2

  • block_size (int) – Size of blocks to convert channels into.

  • layout (string) – One of NCHW or NHWC, indicates channel axis.

  • mode (string) – One of DCR or CDR, indicates which order channels are accessed in.

Returns

result

Tensor with shape [in_batch, in_channel / block_size * block_size,

in_height * block_size, in_width * block_size]

Return type

tvm.relay.Expr

tvm.relay.nn.dilate(data, strides, dilation_value=0.0)

Dilate data with given dilation value (0 by default).

Parameters
  • data (tvm.relay.Expr) – n-D, can be any layout.

  • strides (tuple of <int>) – Dilation stride on each dimension, 1 means no dilation.

  • dilation_value (int/float, optional) – Value used to dilate the input.

Returns

Output – The computed result

Return type

tvm.relay.Expr

tvm.relay.nn.dropout(data, rate=0.5)

Applies the dropout operation to the input array.

During training, each element of the input is set to zero with probability p. The whole array is rescaled by 1/(1-p) to keep the expected sum of the input unchanged.

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • rate (float, optional (default=0.5)) – The probability for an element to be reset to 0.

Returns

result – The result of dropout

Return type

tvm.relay.Expr

tvm.relay.nn.dropout_raw(data, rate=0.5)

Applies the dropout operation to the input array.

During training, each element of the input is set to zero with probability p. The whole array is rescaled by 1/(1-p) to keep the expected sum of the input unchanged.

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • rate (float, optional (default=0.5)) – The probability for an element to be reset to 0.

Returns

result – The result of dropout

Return type

tvm.relay.Expr

tvm.relay.nn.fast_softmax(data, axis=- 1)

Computes softmax. Use approximation to compute exponent for faster speed.

\[\text{softmax}(x)_i = \frac{exp(x_i)}{\sum_j exp(x_j)}\]

Note

This operator can be optimized away for inference.

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • axis (int, optional) – The axis to sum over when computing softmax

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.fifo_buffer(data, buffer, axis)

FIFO buffer to enable computation reuse in CNNs with sliding indow input

Compute equivalent of

concat(buffer, data, axis=axis)
.slice_axis(axis=axis,
            begin=data.shape[axis],
            end=data.shape[axis]+buffer.shape[axis])

Useful for

  • Encoding explicit re-use of computation in convolution ops operated on a sliding window input

  • Implementing a FIFO queue to cache intermediate results, e.g. as in Fast WaveNet.

Parameters
  • data (tvm.relay.Expr) – The input data

  • buffer (tvm.relay.Expr) – Previous value of the FIFO buffer

  • axis (int) – Specify which axis should be used for buffering

Returns

result – Updated value for the buffer

Return type

tvm.relay.Expr

tvm.relay.nn.get_pad_tuple1d(padding)

Common code to get the 1 dimensional pad option :param padding: Padding size :type padding: Union[int, Tuple[int, …]]

Returns

  • pad_left (int) – Padding size on left

  • pad_right (int) – Padding size on right.

tvm.relay.nn.get_pad_tuple2d(padding)

Common code to get the pad option :param padding: Padding size :type padding: Union[int, Tuple[int, …]]

Returns

  • pad_top (int) – Padding size on top

  • pad_left (int) – Padding size on left

  • pad_down (int) – Padding size on down.

  • pad_right (int) – Padding size on right.

tvm.relay.nn.get_pad_tuple3d(padding)

Common code to get the pad option :param padding: Padding size :type padding: Union[int, Tuple[int, …]]

Returns

  • pad_front (int) – Padding size on front

  • pad_top (int) – Padding size on top

  • pad_left (int) – Padding size on left

  • pad_back (int) – Padding size on back

  • pad_down (int) – Padding size on down.

  • pad_right (int) – Padding size on right.

tvm.relay.nn.global_avg_pool1d(data, layout='NCW', out_layout='')

1D global average pooling operator.

This operator takes data as input and does 1D average value calculation across each window represented by W.

In the default case, where the data_layout is NCW a data Tensor with shape (batch_size, in_channels, width), to produce an output Tensor with the following rule:

with data of shape (b, c, w)

\[\mbox{out}(b, c, 1) = \frac{1}{w} \sum_{n=0}^{w-1} \mbox{data}(b, c, n)\]
Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • layout (str, optional) – Layout of the input.

  • out_layout (str, optional) – Layout of the output.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.global_avg_pool2d(data, layout='NCHW', out_layout='')

2D global average pooling operator.

This operator takes data as input and does 2D average value calculation across each window represented by WxH.

In the default case, where the data_layout is NCHW a data Tensor with shape (batch_size, in_channels, height, width), to produce an output Tensor with the following rule:

with data of shape (b, c, h, w)

\[\mbox{out}(b, c, 1, 1) = \frac{1}{h * w} \sum_{m=0}^{h-1} \sum_{n=0}^{w-1} \mbox{data}(b, c, m, n)\]
Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • layout (str, optional) – Layout of the input.

  • out_layout (Optional[str]) – Layout of the output

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.global_avg_pool3d(data, layout='NCDHW', out_layout='')

3D global average pooling operator.

This operator takes data as input and does 3D average value calculation across each window represented by DxWxH.

In the default case, where the data_layout is NCDHW a data Tensor with shape (batch_size, in_channels, depth, height, width), to produce an output Tensor with the following rule:

with data of shape (b, c, d, h, w)

\[\mbox{out}(b, c, 1, 1, 1) = \frac{1}{d * h * w} \sum_{l=0}^{d-1} \sum_{m=0}^{h-1} \sum_{n=0}^{w-1} \mbox{data}(b, c, l, m, n)\]
Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • layout (str, optional) – Layout of the input.

  • out_layout (str, optional) – Layout of the output.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.global_max_pool1d(data, layout='NCW', out_layout='')

1D global maximum pooling operator.

This operator takes data as input and does 1D max value calculation across each window represented by W.

In the default case, where the data_layout is NCW a data Tensor with shape (batch_size, in_channels, width), to produce an output Tensor with the following rule:

with data of shape (b, c, w) .. math:

\mbox{out}(b, c, 1)  = \max_{n=0, \ldots, w} \mbox{data}(b, c, n)
Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • layout (str, optional) – Layout of the input.

  • out_layout (str, optional) – Layout of the output.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.global_max_pool2d(data, layout='NCHW', out_layout='')

2D global maximum pooling operator.

This operator takes data as input and does 2D max value calculation across each window represented by WxH.

In the default case, where the data_layout is NCHW a data Tensor with shape (batch_size, in_channels, height, width), to produce an output Tensor with the following rule:

with data of shape (b, c, h, w)

\[\mbox{out}(b, c, 1, 1) = \max_{m=0, \ldots, h} \max_{n=0, \ldots, w} \mbox{data}(b, c, m, n)\]
Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • layout (str, optional) – Layout of the input.

  • out_layout (Optional[str]) – Layout of the output

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.global_max_pool3d(data, layout='NCDHW', out_layout='')

3D global maximum pooling operator.

This operator takes data as input and does 3D max value calculation across each window represented by DxWxH.

In the default case, where the data_layout is NCDHW a data Tensor with shape (batch_size, in_channels, depth, height, width), to produce an output Tensor with the following rule:

with data of shape (b, c, d, h, w) .. math:

\mbox{out}(b, c, 1, 1, 1)  =  \max_{l=0, \ldots, d},  \max_{m=0, \ldots, h},
     \max_{n=0, \ldots, w} \mbox{data}(b, c, l, m, n)
Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • layout (str, optional) – Layout of the input.

  • out_layout (str, optional) – Layout of the output.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.group_norm(data, gamma, beta, num_groups, axis=1, epsilon=1e-05, center=True, scale=True)

Group normalization normalizes over group of channels for each training examples. We can say that, Group Norm is in between Instance Norm and Layer Norm. When we put all the channels into a single group, group normalization becomes Layer normalization. And, when we put each channel into different groups it becomes Instance normalization

https://arxiv.org/pdf/1803.08494.pdf

Applies group normalization to the n-dimensional input array by seperating the input channels into ‘num_groups’ groups, each containing ‘num_channels / num_groups’ channels. The mean and standard-deviation are calculated separately over the each group. gamma and beta are learnable per-channel affine transform parameter vectors of size num_channels.

\[out = \frac{data - mean(data, axis)}{\sqrt{var(data, axis)+\epsilon}} * gamma + beta\]

Unlike batch normalization, the mean and var are computed along a group of channels.

If the input has size k on axis 1, then both gamma and beta have shape (k,).

Note

This operator can be optimized away for inference.

Parameters
  • data (tvm.relay.Expr) – Input to which group_norm will be applied.

  • gamma (tvm.relay.Expr) – The gamma scale factor.

  • beta (tvm.relay.Expr) – The beta offset factor.

  • num_groups (int) – The number of groups to separate the channels into.

  • axis (int, optional, default=1) – The axis of the channels.

  • epsilon (double, optional, default=1e-5) – Small float added to variance to avoid dividing by zero.

  • center (boolean, optional, default=True) – If True, add offset of beta to normalized tensor, If False, beta is ignored.

  • scale (boolean, optional, default=True) – If True, multiply by gamma. If False, gamma is not used.

Returns

result – The normalized data.

Return type

tvm.relay.Expr

tvm.relay.nn.instance_norm(data, gamma, beta, axis=1, epsilon=1e-05, center=True, scale=True)

Instance Normalization (Ulyanov and et al., 2016) Applies instance normalization to the n-dimensional input array.

\[out = \frac{data - mean(data)}{\sqrt{var(data)+\epsilon}} * gamma + beta\]

The instance normalization is similar to batch normalization, but unlike batch normalization, the mean and var are calculated per-dimension separately for each object(instance) in a mini-batch, not over a batch. And the same normalization is applied both at test and train time.

Assume the input has size k on axis 1, then both gamma and beta have shape (k,).

The parameter axis specifies which axis of the input shape denotes the ‘channel’. The default is 1. Specifying -1 sets the channel axis to be the last item in the input shape.

Note

This operator can be optimized away for inference.

Parameters
  • data (tvm.relay.Expr) – Input to which instance_norm will be applied.

  • gamma (tvm.relay.Expr) – The gamma scale factor.

  • beta (tvm.relay.Expr) – The beta offset factor.

  • axis (int, optional, default=1) – Specify along which shape axis the channel is specified.

  • epsilon (double, optional, default=1e-5) – Small float added to variance to avoid dividing by zero.

  • center (boolean, optional, default=True) – If True, add offset of beta to normalized tensor, If False, beta is ignored.

  • scale (boolean, optional, default=True) – If True, multiply by gamma. If False, gamma is not used.

Returns

  • result (tvm.relay.Expr) – The normalized data.

  • .. _`Instance Normalization (The Missing Ingredient for Fast Stylization`:) – https://arxiv.org/abs/1607.08022

tvm.relay.nn.l2_normalize(data, eps, axis=None)

Perform L2 normalization on the input data

\[y(i, j) = x(i, j) / sqrt(max(sum(x^2), eps))\]
Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • eps (float) – epsilon value

  • axis (list of int, optional) – axis over the normalization applied

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.layer_norm(data, gamma, beta, axis=- 1, epsilon=1e-05, center=True, scale=True)

Layer normalization (Lei Ba and et al., 2016). Applies layer normalization to the n-dimensional input array. This operator takes an n-dimensional input array and normalizes the input using the given axis:

\[out = \frac{data - mean(data, axis)}{\sqrt{var(data, axis)+\epsilon}} * gamma + beta\]

Unlike batch normalization, the mean and var are computed along the channel dimension.

Assume the input has size k on axis 1, then both gamma and beta have shape (k,).

Note

This operator can be optimized away for inference.

Parameters
  • data (tvm.relay.Expr) – Input to which layer_norm will be applied.

  • gamma (tvm.relay.Expr) – The gamma scale factor.

  • beta (tvm.relay.Expr) – The beta offset factor.

  • axis (int, optional, default=-1) – The axis that should be normalized, typically the axis of the channels.

  • epsilon (double, optional, default=1e-5) – Small float added to variance to avoid dividing by zero.

  • center (boolean, optional, default=True) – If True, add offset of beta to normalized tensor, If False, beta is ignored.

  • scale (boolean, optional, default=True) – If True, multiply by gamma. If False, gamma is not used.

Returns

result – The normalized data.

Return type

tvm.relay.Expr

tvm.relay.nn.leaky_relu(data, alpha=0.01)

This operator takes data as input and does Leaky version of a Rectified Linear Unit.

\[`y = x > 0 ? x : alpha * x`\]
Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • alpha (float) – Slope coefficient for the negative half axis.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.log_softmax(data, axis=- 1)

Computes log softmax.

\[\text{log_softmax}(x)_i = \log \frac{exp(x_i)}{\sum_j exp(x_j)}\]

Note

This operator can be optimized away for inference.

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • axis (int, optional) – The axis to sum over when computing log softmax

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.lrn(data, size=5, axis=1, bias=2, alpha=1e-05, beta=0.75)

This operator takes data as input and does local response normalization.

Normalize the input in a local region across or within feature maps. Each input value is divided by (data / (bias + (alpha * sum_data ^2 /size))^beta) where n is the size of each local region, and the sum is taken over the region centered at that value (zero padding is added where necessary).

\[(data / (bias + (alpha * sum_data ^2 /size))^beta)\]
Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • size (int, optional) – The size of the local region to be considered for normalization.

  • axis (int, optional) – Input data layout channel axis. Default value is 1 for NCHW format

  • bias (float, optional) – The offset parameter to avoid dividing by 0.

  • alpha (float, optional) – The scaling parameter.

  • beta (float, optional) – The exponent parameter.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.matmul(tensor_a, tensor_b, units=None, out_dtype='', transpose_a=False, transpose_b=False)

Matmul operator. Applies a linear transformation. The A & B can be transposed.

\[`C = A * B`\]
Parameters
  • data (tvm.relay.Expr) – The first input of the operator, of shape (d_1, d_2, …, d_n, units_in) or (d_1, d_2, …, units_in, d_n).

  • weight (tvm.relay.Expr) – The second input expressions, 2-D matrix, of shape (units_in, units) or (units, units_in).

  • units (Optional[int]) – Number of hidden units of the matmul transformation.

  • out_dtype (Optional[str]) – Specifies the output data type for mixed precision matmul, of shape (d_1, d_2, …, d_n, units).

  • transpose_a (Optional[bool] = False) – Whether the data tensor is in transposed format.

  • transpose_b (Optional[bool] = False) – Whether the weight tensor is in transposed format.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.max_pool1d(data, pool_size=(1,), strides=(1,), dilation=(1,), padding=(0,), layout='NCW', out_layout='', ceil_mode=False)

1D maximum pooling operator.

This operator takes data as input and does 1D max value calculation with in pool_size sized window by striding defined by stride.

In the default case, where the data_layout is NCW a data Tensor with shape (batch_size, channels, width), to produce an output Tensor.

The ceil_mode is used to take ceil or floor while computing out shape. count_include_pad indicates including or excluding padded input values in computation. This operator accepts data layout specification.

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • pool_size (int or tuple of int, optional) – The size of window for pooling.

  • strides (int or tuple of int, optional) – The strides of pooling.

  • dilation (int or tuple of int, optional) – The dilation of pooling.

  • padding (int or tuple of int, optional) – The padding for pooling.

  • layout (str, optional) – Layout of the input.

  • out_layout (Optional[str]) – Layout of the output

  • ceil_mode (bool, optional) – To enable or disable ceil while pooling.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.max_pool2d(data, pool_size=(1, 1), strides=(1, 1), dilation=(1, 1), padding=(0, 0), layout='NCHW', out_layout='', ceil_mode=False)

2D maximum pooling operator.

This operator takes data as input and does 2D max value calculation with in pool_size sized window by striding defined by stride

In the default case, where the data_layout is NCHW a data Tensor with shape (batch_size, in_channels, height, width), to produce an output Tensor with the following rule:

with data of shape (b, c, h, w) and pool_size (kh, kw)

\[\mbox{out}(b, c, y, x) = \max_{m=0, \ldots, kh-1} \max_{n=0, \ldots, kw-1} \mbox{data}(b, c, \mbox{stride}[0] * y + m, \mbox{stride}[1] * x + n)\]

Padding is applied to data before the computation. ceil_mode is used to take ceil or floor while computing out shape. This operator accepts data layout specification.

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • pool_size (int or tuple of int, optional) – The size of window for pooling.

  • strides (tuple of int, optional) – The strides of pooling.

  • dilation (int or tuple of int, optional) – The dilation of pooling.

  • padding (tuple of int, optional) – The padding for pooling.

  • layout (str, optional) – Layout of the input.

  • out_layout (Optional[str]) – Layout of the output

  • ceil_mode (bool, optional) – To enable or disable ceil while pooling.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.max_pool2d_grad(out_grad, data, pool_size=(1, 1), strides=(1, 1), padding=(0, 0), layout='NCHW', out_layout='', ceil_mode=False)

Gradient of 2D maximum pooling operator.

This operator takes out_grad and data as input and calculates gradient of max_pool2d.

Parameters
  • out_grad (tvm.relay.Expr) – The output gradient

  • data (tvm.relay.Expr) – The input data to the operator.

  • pool_size (int or tuple of int, optional) – The size of window for pooling.

  • strides (tuple of int, optional) – The strides of pooling.

  • padding (tuple of int, optional) – The padding for pooling.

  • layout (str, optional) – Layout of the input.

  • out_layout (Optional[str]) – Layout of the output

  • ceil_mode (bool, optional) – To enable or disable ceil while pooling.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.max_pool3d(data, pool_size=(1, 1, 1), strides=(1, 1, 1), dilation=(1, 1, 1), padding=(0, 0, 0), layout='NCDHW', out_layout='', ceil_mode=False)

3D maximum pooling operator.

This operator takes data as input and does 3D max value calculation with in pool_size sized window by striding defined by stride.

In the default case, where the data_layout is NCDHW a data Tensor with shape (batch_size, channels, depth, height, width), to produce an output Tensor.

The ceil_mode is used to take ceil or floor while computing out shape. count_include_pad indicates including or excluding padded input values in computation. This operator accepts data layout specification.

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • pool_size (int or tuple of int, optional) – The size of window for pooling.

  • strides (tuple of int, optional) – The strides of pooling.

  • dilation (int or tuple of int, optional) – The dilation of pooling.

  • padding (tuple of int, optional) – The padding for pooling.

  • layout (str, optional) – Layout of the input.

  • out_layout (Optional[str]) – Layout of the output

  • ceil_mode (bool, optional) – To enable or disable ceil while pooling.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.mirror_pad(data, pad_width, mode='SYMMETRIC')

MirrorPadding

This operator takes in a tensor and pads each axis by the specified widths using mirroring of the border pixels.

Parameters
  • data (tvm.relay.Expr) – The input data to the operator

  • pad_width (tuple of <tuple of <int>>, required) – Number of values padded to the edges of each axis, in the format of ((before_1, after_1), …, (before_N, after_N))

  • mode (string, optional, default='SYMMETRIC') – What type of mirroring to use, must be SYMMETRIC or REFLECT.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.nll_loss(predictions, targets, weights, reduction='mean', ignore_index=- 100)

Negative log likelihood loss.

output{n, i_1, i_2, …, i_k} = -p * w
where t = target{n, i_1, i_2, …, i_k}

p = predictions{n, t, i_1, i_2, i_k} w = weights{n, i_1, i_2, …, i_k} if t != ignore_index else 0

result = reduction(output)

Parameters
  • predictions (tvm.relay.Expr) – The predictions.

  • targets (tvm.relay.Expr) – The target value of each prediction.

  • weights (tvm.relay.Expr) – The weight of each target value.

  • reduction (string) – The reduction method to apply to the output. Possible values are “mean”, “sum” and “none”.

  • ignore_index (int) – The target value to ignore.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.pad(data, pad_width, pad_value=0, pad_mode='constant')

Padding

This operator takes in a tensor and pads each axis by the specified widths using the specified value.

Parameters
  • data (tvm.relay.Expr) – The input data to the operator

  • pad_width (tuple of <tuple of <int>>, or tvm.relay.Expr, required) – Number of values padded to the edges of each axis, in the format of ((before_1, after_1), …, (before_N, after_N))

  • pad_value (float, or tvm.relay.Expr, optional, default=0) – The value used for padding

  • pad_mode ('constant', 'edge', 'reflect') – ‘constant’ pads with constant_value pad_value ‘edge’ pads using the edge values of the input array ‘reflect’ pads by reflecting values with respect to the edge

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.prelu(data, alpha, axis=1)

This operator takes data as input and does Leaky version of a Rectified Linear Unit.

\[y = x > 0 ? x : alpha * x\]
Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • alpha (tvm.relay.Expr) – Slope coefficient for the negative half axis.

  • axis (int, optional) – Specify which shape axis the channel is specified.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.relu(data)

Rectified linear unit.

\[out = max(x, 0)\]
Parameters

data (tvm.relay.Expr) – The input data

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.softmax(data, axis=- 1)

Computes softmax.

\[\text{softmax}(x)_i = \frac{exp(x_i)}{\sum_j exp(x_j)}\]

Note

This operator can be optimized away for inference.

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • axis (int, optional) – The axis to sum over when computing softmax

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.space_to_batch_nd(data, block_shape, paddings, pad_value=0)

Divide spatial dimensions of the data into a grid of blocks and interleave them into batch dim.

Parameters
  • data (tvm.te.Tensor) – N-D with shape [batch, spatial_shape, remaining_shape]

  • block_shape (relay.Expr) – 1-D of size [M] where M is number of spatial dims, specifies block size for each spatial dimension.

  • paddings (relay.Expr) – 2-D of shape [M, 2] where M is number of spatial dims, specifies [before, after] paddings for each spatial dimension.

  • pad_value (float, or relay.Expr, optional, default=0) – The value used for padding.

Returns

result – N-D Tensor with shape [in_batch * prod(block_shape), padded_data[1] / block_shape[0], …, padded_data[M] / block_shape[M-1], remaining_shape]

Return type

relay.Expr

tvm.relay.nn.space_to_depth(data, block_size, layout='NCHW')

Convert spatial blocks into channels.

Parameters
  • data (tvm.relay.Expr) – Input data with spatial dimensions divisible by block_size

  • block_size (int) – Size of blocks to decompose into channels.

  • layout (string) – One of NCHW or NHWC, indicates channel axis.

Returns

result

Tensor with shape [in_batch, in_channel * block_size * block_size,

in_height / block_size, in_width / block_size]

Return type

tvm.relay.Expr

tvm.relay.nn.sparse_add(dense_mat, sparse_mat)

Computes the matrix addition of dense_mat and sparse_mat, where dense_mat is a dense matrix and sparse_mat is a sparse (CSR) namedtuple with fields data, indices, and indptr.

\[\mbox{sparse_add}(dense_mat, sparse_mat)[m, n] = \mbox{add}(\mbox{as_dense}(S), (D))[m, n]\]

where as_dense returns dense equivalent of the given S(sparse matrix) while performing addition with given D(dense matrix).

Parameters
  • dense_mat (tvm.relay.Expr) – The input dense matrix for the matrix addition

  • sparse_mat (Union[namedtuple, Tuple[ndarray, ndarray, ndarray]]) – The input sparse matrix(CSR) for the matrix addition.

Returns

result – The computed result.

Return type

tvm.relay.Expr

Examples

dense_data = [[ 3.,   4.,   4. ]
              [ 4.,  2.,  5. ]]
sparse_data = [4., 8.]
sparse_indices =[0, 2]
sparse_indptr =[0, 1, 2]

output = relay.sparse_add(dense_data, sparse_data, sparse_indices, sparse_indptr)

output = [[ 7.,   4.,   4. ]
          [ 4.,  2.,  13. ]]
tvm.relay.nn.sparse_dense(dense_mat, sparse_mat, sparse_lhs=False)

Computes the matrix multiplication of dense_mat and sparse_mat, where dense_mat is a dense matrix and sparse_mat is a sparse (either BSR or CSR) namedtuple with fields data, indices, and indptr.

if sparse_lhs=False:
\[\mbox{sparse_dense}(dense_mat, sparse_mat)[m, n] = \mbox{matmul}(D, \mbox{as_dense}(S)^T)[m, n]\]
if sparse_lhs=True:
\[\mbox{sparse_dense}(dense_mat, sparse_mat)[m, n] = \mbox{matmul}(\mbox{as_dense}(S), (D)^T)[m, n]\]

where as_dense returns dense equivalent of the given S(sparse matrix) while performing matmul with given D(dense matrix).

See https://docs.scipy.org/doc/scipy/reference/generated/scipy.sparse.csr_matrix.html and https://docs.scipy.org/doc/scipy-0.14.0/reference/generated/scipy.sparse.bsr_matrix.html for more detail on the sparse matrix representation.

Parameters
  • dense_mat (tvm.relay.Expr) – The input dense matrix for the matrix multiplication

  • sparse_mat (Union[namedtuple, Tuple[ndarray, ndarray, ndarray]]) – The input sparse matrix for the matrix multiplication.

  • sparse_lhs (bool, optional) – Indicates whether lhs or rhs matrix is sparse. Default value is False.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.sparse_transpose(x)

Computes the fast matrix transpose of x, where x is a sparse tensor in CSR format (represented as a namedtuple with fields data, indices, and indptr).

** Currently only support Square Matrices **

\[\mbox{sparse_transpose}(x)[n, n] = (x^T)[n, n]\]

Please refer to https://github.com/scipy/scipy/blob/v1.3.0/scipy/sparse/csr.py for the algorithm implemented in this operator.

Parameters

x (Union[namedtuple, Tuple[ndarray, ndarray, ndarray]]) – The sparse weight matrix for the fast matrix transpose.

Returns

result – Tuple of output sparse tensor (same shape and format as input), i.e. if CSR then output is in ([data, indices, indptr]) form

Return type

relay.Tuple([tvm.relay.Expr, tvm.relay.Expr, tvm.relay.Expr])

tvm.relay.nn.upsampling(data, scale_h=1, scale_w=1, layout='NCHW', method='nearest_neighbor', align_corners=False)

Upsampling.

This operator takes data as input and does 2D scaling to the given scale factor. In the default case, where the data_layout is NCHW with data of shape (n, c, h, w) out will have a shape (n, c, h*scale_h, w*scale_w)

method indicates the algorithm to be used while calculating the out value and method can be one of (“bilinear”, “nearest_neighbor”, “bicubic”)

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • scale_h (tvm.relay.Expr or int or float) – The scale factor for height upsampling.

  • scale_w (tvm.relay.Expr or int or float) – The scale factor for width upsampling.

  • layout (str, optional) – Layout of the input.

  • method (str, optional) – Scale method to used [nearest_neighbor, bilinear, bicubic].

  • align_corners (bool, optional) – Whether to keep corners in proper place.

Returns

result – The computed result.

Return type

tvm.relay.Expr

tvm.relay.nn.upsampling3d(data, scale_d=1, scale_h=1, scale_w=1, layout='NCDHW', method='nearest_neighbor', coordinate_transformation_mode='half_pixel')

3D Upsampling.

This operator takes data as input and does 3D scaling to the given scale factor. In the default case, where the data_layout is NCDHW with data of shape (n, c, d, h, w) out will have a shape (n, c, d*scale_d, h*scale_h, w*scale_w)

method indicates the algorithm to be used while calculating the out value and method can be one of (“trilinear”, “nearest_neighbor”)

Parameters
  • data (tvm.relay.Expr) – The input data to the operator.

  • scale_d (tvm.relay.Expr) – The scale factor for depth upsampling.

  • scale_h (tvm.relay.Expr) – The scale factor for height upsampling.

  • scale_w (tvm.relay.Expr) – The scale factor for width upsampling.

  • layout (str, optional) – Layout of the input.

  • method (str, optional) – Scale method to used [nearest_neighbor, trilinear].

  • coordinate_transformation_mode (string, optional) – Describes how to transform the coordinate in the resized tensor to the coordinate in the original tensor. Refer to the ONNX Resize operator specification for details. Available options are “half_pixel”, “align_corners” and “asymmetric”.

Returns

result – The computed result.

Return type

tvm.relay.Expr