Skip to content

Commit

Permalink
Merge pull request #332 from ModECI/development
Browse files Browse the repository at this point in the history 
Release 0.4.3
  • Loading branch information
@pgleeson
pgleeson authored Oct 6, 2022
2 parents 032aa1c + ba11ea8 commit a2f3ea9
Show file tree
Hide file tree
Showing 71 changed files with 67,913 additions and 499 deletions.
2 changes: 1 addition & 1 deletion .pre-commit-config.yaml
View file
Original file line number Diff line number Diff line change
Expand Up @@ -9,7 +9,7 @@ repos:
rev: v3.4.0
hooks:
- id: check-added-large-files
args: ['--maxkb=800']
args: ['--maxkb=3000']
- id: check-case-conflict
- id: check-merge-conflict
- id: check-symlinks
Expand Down
41 changes: 28 additions & 13 deletions docs/MDF_function_specifications.json
View file
Original file line number Diff line number Diff line change
Expand Up @@ -92,6 +92,13 @@
],
"expression_string": "A * (A > 0)"
},
"onnx::HardSwish": {
"description": "\nHardSwish takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where\nthe HardSwish function, y = x * max(0, min(1, alpha * x + beta)) = x * HardSigmoid<alpha, beta>(x),\nwhere alpha = 1/6 and beta = 0.5, is applied to the tensor elementwise.\n",
"arguments": [
"X"
],
"expression_string": "onnx_ops.hardswish(X)"
},
"onnx::LessOrEqual": {
"description": "\nReturns the tensor resulted from performing the `less_equal` logical operation\nelementwise on the input tensors `A` and `B` (with Numpy-style broadcasting support).\n\nThis operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md).\n",
"arguments": [
Expand Down Expand Up @@ -210,6 +217,14 @@
],
"expression_string": "onnx_ops.bitshift(X, Y, direction)"
},
"onnx::Trilu": {
"description": "\nGiven a 2-D matrix or batches of 2-D matrices, returns the upper or lower triangular part of the tensor(s).\nThe attribute \"upper\" determines whether the upper or lower part is retained. If set to true,\nthe upper triangular matrix is retained. Lower triangular matrix is retained otherwise.\nDefault value for the \"upper\" attribute is true.\nTrilu takes one input tensor of shape [*, N, M], where * is zero or more batch dimensions. The upper triangular part consists\nof the elements on and above the given diagonal (k). The lower triangular part consists of elements on and below the diagonal.\nAll other elements in the matrix are set to zero.\nIf k = 0, the triangular part on and above/below the main diagonal is retained.\nIf upper is set to true, a positive k retains the upper triangular matrix excluding the main diagonal and (k-1) diagonals above it.\nA negative k value retains the main diagonal and |k| diagonals below it.\nIf upper is set to false, a positive k retains the lower triangular matrix including the main diagonal and k diagonals above it.\nA negative k value excludes the main diagonal and (|k|-1) diagonals below it.\n",
"arguments": [
"input",
"k"
],
"expression_string": "onnx_ops.trilu(input, k, upper)"
},
"onnx::RoiAlign": {
"description": "\nRegion of Interest (RoI) align operation described in the\n[Mask R-CNN paper](https://arxiv.org/abs/1703.06870).\nRoiAlign consumes an input tensor X and region of interests (rois)\nto apply pooling across each RoI; it produces a 4-D tensor of shape\n(num_rois, C, output_height, output_width).\n\nRoiAlign is proposed to avoid the misalignment by removing\nquantizations while converting from original image into feature\nmap and from feature map into RoI feature; in each ROI bin,\nthe value of the sampled locations are computed directly\nthrough bilinear interpolation.\n",
"arguments": [
Expand Down Expand Up @@ -533,7 +548,7 @@
"expression_string": "onnx_ops.tile(input, repeats)"
},
"onnx::Sub": {
"description": "\nPerforms element-wise binary subtraction (with Numpy-style broadcasting support).\n\nThis operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md).\n",
"description": "\nPerforms element-wise binary subtraction (with Numpy-style broadcasting support).\n\nThis operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md).\n\n(Opset 14 change): Extend supported types to include uint8, int8, uint16, and int16.\n",
"arguments": [
"A",
"B"
Expand Down Expand Up @@ -730,7 +745,7 @@
"expression_string": "onnx_ops.reciprocal(X)"
},
"onnx::Mul": {
"description": "\nPerforms element-wise binary multiplication (with Numpy-style broadcasting support).\n\nThis operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md).\n",
"description": "\nPerforms element-wise binary multiplication (with Numpy-style broadcasting support).\n\nThis operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md).\n\n(Opset 14 change): Extend supported types to include uint8, int8, uint16, and int16.\n",
"arguments": [
"A",
"B"
Expand Down Expand Up @@ -782,7 +797,7 @@
"sequence_lens",
"initial_h"
],
"expression_string": "onnx_ops.rnn(X, W, R, B, sequence_lens, initial_h, activation_alpha, activation_beta, activations, clip, direction, hidden_size)"
"expression_string": "onnx_ops.rnn(X, W, R, B, sequence_lens, initial_h, activation_alpha, activation_beta, activations, clip, direction, hidden_size, layout)"
},
"onnx::Pad": {
"description": "\nGiven a tensor containing the data to be padded (`data`), a tensor containing the number of start and end pad values for axis (`pads`), (optionally) a `mode`, and (optionally) `constant_value`,\na padded tensor (`output`) is generated.\n\nThe three supported `modes` are (similar to corresponding modes supported by `numpy.pad`):\n\n1) `constant`(default) - pads with a given constant value as specified by `constant_value` (which defaults to 0, empty string, or False)\n\n2) `reflect` - pads with the reflection of the vector mirrored on the first and last values of the vector along each axis\n\n3) `edge` - pads with the edge values of array\n\n\nExample 1 (`constant` mode):\n Insert 0 pads to the beginning of the second dimension.\n\n data =\n [\n [1.0, 1.2],\n [2.3, 3.4],\n [4.5, 5.7],\n ]\n\n pads = [0, 2, 0, 0]\n\n mode = 'constant'\n\n constant_value = 0.0\n\n output =\n [\n [0.0, 0.0, 1.0, 1.2],\n [0.0, 0.0, 2.3, 3.4],\n [0.0, 0.0, 4.5, 5.7],\n ]\n\n\nExample 2 (`reflect` mode):\n data =\n [\n [1.0, 1.2],\n [2.3, 3.4],\n [4.5, 5.7],\n ]\n\n pads = [0, 2, 0, 0]\n\n mode = 'reflect'\n\n output =\n [\n [1.0, 1.2, 1.0, 1.2],\n [2.3, 3.4, 2.3, 3.4],\n [4.5, 5.7, 4.5, 5.7],\n ]\n\n\nExample 3 (`edge` mode):\n data =\n [\n [1.0, 1.2],\n [2.3, 3.4],\n [4.5, 5.7],\n ]\n\n pads = [0, 2, 0, 0]\n\n mode = 'edge'\n\n output =\n [\n [1.0, 1.0, 1.0, 1.2],\n [2.3, 2.3, 2.3, 3.4],\n [4.5, 4.5, 4.5, 5.7],\n ]\n\n",
Expand Down Expand Up @@ -843,12 +858,12 @@
"expression_string": "onnx_ops.mean(data_0)"
},
"onnx::Reshape": {
"description": "\nReshape the input tensor similar to numpy.reshape.\nFirst input is the data tensor, second input is a shape tensor which specifies the output shape. It outputs the reshaped tensor.\nAt most one dimension of the new shape can be -1. In this case, the value is\ninferred from the size of the tensor and the remaining dimensions. A dimension\ncould also be 0, in which case the actual dimension value is unchanged (i.e. taken\nfrom the input tensor).",
"description": "\nReshape the input tensor similar to numpy.reshape.\nFirst input is the data tensor, second input is a shape tensor which specifies the output shape. It outputs the reshaped tensor.\nAt most one dimension of the new shape can be -1. In this case, the value is\ninferred from the size of the tensor and the remaining dimensions. A dimension\ncould also be 0, in which case the actual dimension value is unchanged (i.e. taken\nfrom the input tensor). If 'allowzero' is set, and the new shape includes 0, the\ndimension will be set explicitly to zero (i.e. not taken from input tensor)",
"arguments": [
"data",
"shape"
],
"expression_string": "onnx_ops.reshape(data, shape)"
"expression_string": "onnx_ops.reshape(data, shape, allowzero)"
},
"onnx::ReduceL2": {
"description": "\nComputes the L2 norm of the input tensor's element along the provided axes. The resulted\ntensor has the same rank as the input if keepdims equal 1. If keepdims equal 0, then\nthe resulted tensor have the reduced dimension pruned.\n\nThe above behavior is similar to numpy, with the exception that numpy default keepdims to\nFalse instead of True.",
Expand Down Expand Up @@ -923,15 +938,15 @@
"expression_string": "onnx_ops.leakyrelu(X, alpha)"
},
"onnx::BatchNormalization": {
"description": "\nCarries out batch normalization as described in the paper\nhttps://arxiv.org/abs/1502.03167. Depending on the mode it is being run,\nthere are multiple cases for the number of outputs, which we list below:\n\nOutput case #1: Y, mean, var, saved_mean, saved_var (training mode)\nOutput case #2: Y (test mode)\n\nFor previous (depreciated) non-spatial cases, implementors are suggested\nto flatten the input shape to (N x C*D1*D2 ..*Dn) before a BatchNormalization Op.\nThis operator has **optional** inputs/outputs. See [the doc](IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted.\n",
"description": "\nCarries out batch normalization as described in the paper\nhttps://arxiv.org/abs/1502.03167. Depending on the mode it is being run,\nThere are five required inputs 'X', 'scale', 'B', 'input_mean' and\n'input_var'.\nNote that 'input_mean' and 'input_var' are expected to be the estimated\nstatistics in inference mode (training_mode=False, default),\nand the running statistics in training mode (training_mode=True).\nThere are multiple cases for the number of outputs, which we list below:\n\nOutput case #1: Y, running_mean, running_var (training_mode=True)\nOutput case #2: Y (training_mode=False)\n\nWhen training_mode=False, extra outputs are invalid.\nThe outputs are updated as follows when training_mode=True:\n```\nrunning_mean = input_mean * momentum + current_mean * (1 - momentum)\nrunning_var = input_var * momentum + current_var * (1 - momentum)\n\nY = (X - current_mean) / sqrt(current_var + epsilon) * scale + B\n\nwhere:\n\ncurrent_mean = ReduceMean(X, axis=all_except_channel_index)\ncurrent_var = ReduceVar(X, axis=all_except_channel_index)\n\nNotice that ReduceVar refers to the population variance, and it equals to\nsum(sqrd(x_i - x_avg)) / N\nwhere N is the population size (this formula does not use sample size N - 1).\n\n```\n\nWhen training_mode=False:\n```\nY = (X - input_mean) / sqrt(input_var + epsilon) * scale + B\n```\n\nFor previous (depreciated) non-spatial cases, implementors are suggested\nto flatten the input shape to (N x C * D1 * D2 * ... * Dn) before a BatchNormalization Op.\nThis operator has **optional** inputs/outputs. See [the doc](IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted.\n",
"arguments": [
"X",
"scale",
"B",
"mean",
"var"
"input_mean",
"input_var"
],
"expression_string": "onnx_ops.batchnormalization(X, scale, B, mean, var, epsilon, momentum)"
"expression_string": "onnx_ops.batchnormalization(X, scale, B, input_mean, input_var, epsilon, momentum, training_mode)"
},
"onnx::Cosh": {
"description": "\nCalculates the hyperbolic cosine of the given input tensor element-wise.\n",
Expand Down Expand Up @@ -967,7 +982,7 @@
"initial_c",
"P"
],
"expression_string": "onnx_ops.lstm(X, W, R, B, sequence_lens, initial_h, initial_c, P, activation_alpha, activation_beta, activations, clip, direction, hidden_size, input_forget)"
"expression_string": "onnx_ops.lstm(X, W, R, B, sequence_lens, initial_h, initial_c, P, activation_alpha, activation_beta, activations, clip, direction, hidden_size, input_forget, layout)"
},
"onnx::Unsqueeze": {
"description": "\nInsert single-dimensional entries to the shape of an input tensor (`data`).\nTakes one required input `axes` - which contains a list of dimension indices and this operator will insert a dimension of value `1` into the corresponding index of the output tensor (`expanded`).\n\nFor example:\n Given an input tensor (`data`) of shape [3, 4, 5], then\n Unsqueeze(data, axes=[0, 4]) outputs a tensor (`expanded`) containing same data as `data` but with shape [1, 3, 4, 5, 1].\n\nThe input `axes` should not contain any duplicate entries. It is an error if it contains duplicates.\nThe rank of the output tensor (`output_rank`) is the rank of the input tensor (`data`) plus the number of values in `axes`.\nEach value in `axes` should be within the (inclusive) range [-output_rank , output_rank - 1].\nThe order of values in `axes` does not matter and can come in any order.\n\n",
Expand Down Expand Up @@ -1128,7 +1143,7 @@
"sequence_lens",
"initial_h"
],
"expression_string": "onnx_ops.gru(X, W, R, B, sequence_lens, initial_h, activation_alpha, activation_beta, activations, clip, direction, hidden_size, linear_before_reset)"
"expression_string": "onnx_ops.gru(X, W, R, B, sequence_lens, initial_h, activation_alpha, activation_beta, activations, clip, direction, hidden_size, layout, linear_before_reset)"
},
"onnx::Resize": {
"description": "\nResize the input tensor. In general, it calculates every value in the output tensor as a weighted average of neighborhood (a.k.a. sampling locations) in the input tensor.\nEach dimension value of the output tensor is:\n output_dimension = floor(input_dimension * (roi_end - roi_start) * scale) if input \\\"sizes\\\" is not specified.\n",
Expand Down Expand Up @@ -1197,7 +1212,7 @@
"expression_string": "onnx_ops.argmin(data, axis, keepdims, select_last_index)"
},
"onnx::Add": {
"description": "\nPerforms element-wise binary addition (with Numpy-style broadcasting support).\n\nThis operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md).\n",
"description": "\nPerforms element-wise binary addition (with Numpy-style broadcasting support).\n\nThis operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md).\n\n(Opset 14 change): Extend supported types to include uint8, int8, uint16, and int16.\n",
"arguments": [
"A",
"B"
Expand Down Expand Up @@ -1248,7 +1263,7 @@
"expression_string": "onnx_ops.min(data_0)"
},
"onnx::Div": {
"description": "\nPerforms element-wise binary division (with Numpy-style broadcasting support).\n\nThis operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md).\n",
"description": "\nPerforms element-wise binary division (with Numpy-style broadcasting support).\n\nThis operator supports **multidirectional (i.e., Numpy-style) broadcasting**; for more details please check [the doc](Broadcasting.md).\n\n(Opset 14 change): Extend supported types to include uint8, int8, uint16, and int16.\n",
"arguments": [
"A",
"B"
Expand Down
Loading

0 comments on commit a2f3ea9

Please sign in to comment.