From 28eb7b6fe5673ff511b6514fd9003db03e08226d Mon Sep 17 00:00:00 2001 From: Travis CI Date: Sun, 14 Jan 2018 09:25:37 +0000 Subject: [PATCH] Deploy to GitHub Pages: 5ad1aef051349a73b00b8d611f0ae2508f02490b --- develop/doc/operators.json | 1816 ++++++++--------- .../_sources/howto/dev/new_op_cn.md.txt | 2 +- develop/doc_cn/howto/dev/new_op_cn.html | 2 +- 3 files changed, 813 insertions(+), 1007 deletions(-) diff --git a/develop/doc/operators.json b/develop/doc/operators.json index 62438e944d..e7524b5801 100644 --- a/develop/doc/operators.json +++ b/develop/doc/operators.json @@ -119,182 +119,6 @@ "comment" : "(float, default 1.0e-6) Constant for numerical stability", "generated" : 0 } ] -},{ - "type" : "conv3d", - "comment" : "\nConvolution3D Operator.\n\nThe convolution operation calculates the output based on the input, filter\nand strides, paddings, dilations, groups parameters. The size of each dimension of the\nparameters is checked in the infer-shape.\nInput(Input) and output(Output) are in NCDHW format, where N is batch\nsize, C is the number of channels,D is the depth of the feature, H is the height of\nthe feature, and W is the width of the feature.\nFilters(Input) is MCDHW format, where M is the number of output image channels,\nC is the number of input image channels, D is the depth of the filter,\nH is the height of the filter, and W is the width of the filter.\nParameters(strides, paddings, dilations) are three elements. These three elements\nrepresent depth, height and width, respectively.\nThe input(X) size and output(Out) size may be different.\n\nExample:\n Input:\n Input shape: $(N, C_{in}, D_{in}, H_{in}, W_{in})$\n Filter shape: $(C_{out}, C_{in}, D_f, H_f, W_f)$\n Output:\n Output shape: $(N, C_{out}, D_{out}, H_{out}, W_{out})$\n Where\n $$\n D_{out}= \\frac{(D_{in} + 2 * paddings[0] - (dilations[0] * (D_f - 1) + 1))}{ strides[0]}+ 1 \\\\\n H_{out}= \\frac{(H_{in} + 2 * paddings[1] - (dilations[1] * (H_f - 1) + 1))}{ strides[1]}+ 1 \\\\\n W_{out}= \\frac{(W_{in} + 2 * paddings[2] - (dilations[2] * (W_f - 1) + 1))}{ strides[2]}+ 1\n $$\n", - "inputs" : [ - { - "name" : "Input", - "comment" : "(Tensor) The input tensor of convolution operator. The format of input tensor is NCDHW. Where N is batch size, C is the number of channels, D is the depth of the feature, H is the height of the feature, and W is the width of the feature.", - "duplicable" : 0, - "intermediate" : 0 - }, { - "name" : "Filter", - "comment" : "(Tensor) The filter tensor of convolution operator. The format of the filter tensor is MCDHW, where M is the number of output image channels, C is the number of input image channels, D is the depth of the filter, H is the height of the filter, and W is the width of the filter.If the groups attribute is greater than 1, C equals the number of input image channels divided by the groups.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "outputs" : [ - { - "name" : "Output", - "comment" : "(Tensor) The output tensor of convolution operator.The format of output tensor is also NCDHW.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "attrs" : [ - { - "name" : "strides", - "type" : "int array", - "comment" : "(vector, default:{1, 1, 1}), the strides(d_stride, h_stride, w_stride) of convolution operator.", - "generated" : 0 - }, { - "name" : "paddings", - "type" : "int array", - "comment" : "(vector, default:{0, 0, 0}), the paddings(d_pad, h_pad, w_pad) of convolution operator.", - "generated" : 0 - }, { - "name" : "groups", - "type" : "int", - "comment" : "(int default:1), the groups number of the convolution operator. According to grouped convolution in Alex Krizhevsky's Deep CNN paper: when group=2, the first half of the filters is only connected to the first half of the input channels, while the second half of the filters is only connected to the second half of the input channels.", - "generated" : 0 - }, { - "name" : "dilations", - "type" : "int array", - "comment" : "(vector default:{1, 1, 1}), the dilations(d_dilation, h_dilation, w_dilation) of convolution operator.", - "generated" : 0 - } ] -},{ - "type" : "conv2d", - "comment" : "\nConvolution Operator.\n\nThe convolution operation calculates the output based on the input, filter\nand strides, paddings, dilations, groups parameters. The size of each dimension of the\nparameters is checked in the infer-shape.\nInput(Input) and Output(Output) are in NCHW format. Where N is batch\nsize, C is the number of channels, H is the height of the feature, and W is\nthe width of the feature.\nFilters(Input) is MCHW format. Where M is the number of output image channels, C is\nthe number of input image channels, H is the height of the filter, and W\nis the width of the filter.\nParameters(strides, paddings, dilations) are two elements. These two elements represent\nheight and width, respectively.\nThe input(X) size and output(Out) size may be different.\n\nExample:\n Input:\n Input shape: $(N, C_{in}, H_{in}, W_{in})$\n Filter shape: $(C_{out}, C_{in}, H_f, W_f)$\n Output:\n Output shape: $(N, C_{out}, H_{out}, W_{out})$\n Where\n$$\n H_{out}= \\frac{(H_{in} + 2 * paddings[0] - (dilations[0] * (H_f - 1) + 1))}{strides[0]}+ 1 \\\\\n W_{out}= \\frac{(W_{in} + 2 * paddings[1] - (dilations[1] * (W_f - 1) + 1))}{strides[1]}+ 1\n$$\n", - "inputs" : [ - { - "name" : "Input", - "comment" : "(Tensor) The input tensor of convolution operator. The format of input tensor is NCHW, where N is batch size, C is the number of channels, H is the height of the feature, and W is the width of the feature.", - "duplicable" : 0, - "intermediate" : 0 - }, { - "name" : "Filter", - "comment" : "(Tensor) The filter tensor of convolution operator. The format of the filter tensor is MCHW, where M is the number of output image channels, C is the number of input image channels, H is the height of the filter, and W is the width of the filter. If the groups attribute is greater than 1, C equals the number of input image channels divided by the groups.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "outputs" : [ - { - "name" : "Output", - "comment" : "(Tensor) The output tensor of convolution operator. The format of output tensor is also NCHW.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "attrs" : [ - { - "name" : "strides", - "type" : "int array", - "comment" : "(vector default:{1, 1}), the strides(h_stride, w_stride) of convolution operator.", - "generated" : 0 - }, { - "name" : "paddings", - "type" : "int array", - "comment" : "(vector default:{0, 0}), the paddings(h_pad, w_pad) of convolution operator.", - "generated" : 0 - }, { - "name" : "groups", - "type" : "int", - "comment" : "(int default:1), the groups number of the convolution operator. According to grouped convolution in Alex Krizhevsky's Deep CNN paper: when group=2, the first half of the filters is only connected to the first half of the input channels, while the second half of the filters is only connected to the second half of the input channels.", - "generated" : 0 - }, { - "name" : "dilations", - "type" : "int array", - "comment" : "(vector default:{1, 1}), the dilations(h_dilation, w_dilation) of convolution operator.", - "generated" : 0 - } ] -},{ - "type" : "pool3d", - "comment" : "\nPool3d Operator.\n\nThe pooling3d operation calculates the output based on\nthe input, pooling_type, ksize, strides, and paddings parameters.\nInput(X) and output(Out) are in NCDHW format, where N is batch\nsize, C is the number of channels, and D, H and W are the depth, height and\nwidth of the feature, respectively. Parameters(ksize, strides, paddings) \nare three elements. These three elements represent depth, height and \nwidth, respectively. The input(X) size and output(Out) size may be different.\n\nExample:\n Input:\n X shape: $(N, C, D_{in}, H_{in}, W_{in})$\n Output:\n Out shape: $(N, C, D_{out}, H_{out}, W_{out})$\n Where\n $$\n D_{out} = \\frac{(D_{in} - ksize[0] + 2 * paddings[0])}{strides[0]} + 1 \\\\\n H_{out} = \\frac{(H_{in} - ksize[1] + 2 * paddings[1])}{strides[1]} + 1 \\\\\n W_{out} = \\frac{(W_{in} - ksize[2] + 2 * paddings[2])}{strides[2]} + 1\n $$\n\n", - "inputs" : [ - { - "name" : "X", - "comment" : "(Tensor) The input tensor of pooling operator. The format of input tensor is NCDHW, where N is batch size, C is the number of channels, and D, H and W is the depth, height and width of the feature, respectively.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "outputs" : [ - { - "name" : "Out", - "comment" : "(Tensor) The output tensor of pooling operator.The format of output tensor is also NCDHW, where N is batch size, C is the number of channels, and D, H and W is the depth, height and width of the feature, respectively.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "attrs" : [ - { - "name" : "pooling_type", - "type" : "string", - "comment" : "(string) Pooling type, can be \"max\" for max-pooling and \"avg\" for average-pooling.", - "generated" : 0 - }, { - "name" : "ksize", - "type" : "int array", - "comment" : "(vector) The pooling window size(depth, height, width) of pooling operator. If global_pooling = true, ksize and paddings will be ignored.", - "generated" : 0 - }, { - "name" : "global_pooling", - "type" : "bool", - "comment" : "(bool, default false) Whether to use the global pooling. If global_pooling = true, ksize and paddings wille be ignored.", - "generated" : 0 - }, { - "name" : "strides", - "type" : "int array", - "comment" : "(vector, default {1,1,1}) Strides(depth, height, width) of the pooling operator.", - "generated" : 0 - }, { - "name" : "paddings", - "type" : "int array", - "comment" : "(vector, default {0,0,0}), paddings(depth, height, width) of pooling operator. If global_pooling = true, ksize and paddings will be ignored.", - "generated" : 0 - } ] -},{ - "type" : "pool2d", - "comment" : "\nPool2d Operator.\n\nThe pooling2d operation calculates the output based on\nthe input, pooling_type and ksize, strides, paddings parameters.\nInput(X) and output(Out) are in NCHW format, where N is batch size, C is the\nnumber of channels, H is the height of the feature, and W is the width of the feature.\nParameters(ksize, strides, paddings) are two elements.\nThese two elements represent height and width, respectively.\nThe input(X) size and output(Out) size may be different.\n\nExample: \n Input:\n X shape: $(N, C, H_{in}, W_{in})$\n Output:\n Out shape: $(N, C, H_{out}, W_{out})$\n Where\n $$ \n H_{out} = \\frac{(H_{in} - ksize[0] + 2 * paddings[0])}{strides[0]} + 1 \\\\\n W_{out} = \\frac{(W_{in} - ksize[1] + 2 * paddings[1])}{strides[1]} + 1\n $$\n\n", - "inputs" : [ - { - "name" : "X", - "comment" : "(Tensor) The input tensor of pooling operator. The format of input tensor is NCHW, where N is batch size, C is the number of channels, H is the height of the feature, and W is the width of the feature.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "outputs" : [ - { - "name" : "Out", - "comment" : "(Tensor) The output tensor of pooling operator. The format of output tensor is also NCHW, where N is batch size, C is the number of channels, H is the height of the feature, and W is the width of the feature.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "attrs" : [ - { - "name" : "pooling_type", - "type" : "string", - "comment" : "(string), pooling type, can be \"max\" for max-pooling and \"avg\" for average-pooling.", - "generated" : 0 - }, { - "name" : "ksize", - "type" : "int array", - "comment" : "(vector) The pooling window size(height, width) of the pooling operator. If global_pooling = true, ksize and paddings will be ignored.", - "generated" : 0 - }, { - "name" : "global_pooling", - "type" : "bool", - "comment" : "(bool, default false) Whether to use the global pooling. If global_pooling = true, ksize and paddings will be ignored.", - "generated" : 0 - }, { - "name" : "strides", - "type" : "int array", - "comment" : "(vector, default {1, 1}), strides(height, width) of pooling operator.", - "generated" : 0 - }, { - "name" : "paddings", - "type" : "int array", - "comment" : "(vector, default {0,0}), paddings(height, width) of pooling operator.If global_pooling = true, paddings and ksize will be ignored.", - "generated" : 0 - } ] },{ "type" : "max_pool3d_with_index", "comment" : "\nMaxPool3d Operator.\n\nThe maxpooling3d with index operation calculates the output and the mask\nbased on the input and ksize, strides, paddings parameters.\nInput(X) and output(Out, Mask) are in NCDHW format, where N is batch\nsize, C is the number of channels, and D, H and W are the depth, height and\nwidth of the feature, respectively. \nParameters(ksize, strides, paddings) are three elements.\nThese three elements represent depth, height and width, respectively.\nThe input(X) size and output(Out, Mask) size may be different.\n\nExample:\n Input:\n X shape: $(N, C, D_{in}, H_{in}, W_{in})$\n Output:\n Out shape: $(N, C, D_{out}, H_{out}, W_{out})$\n Mask shape: $(N, C, D_{out}, H_{out}, W_{out})$\n Where\n $$\n D_{out} = \\frac{(D_{in} - ksize[0] + 2 * paddings[0])}{strides[0]} + 1 \\\\\n H_{out} = \\frac{(H_{in} - ksize[1] + 2 * paddings[1])}{strides[1]} + 1 \\\\\n W_{out} = \\frac{(W_{in} - ksize[2] + 2 * paddings[2])}{strides[2]} + 1\n $$\n\n", @@ -514,45 +338,6 @@ "comment" : "(string, default: tanh)The activation for candidate hidden state, `tanh` by default.", "generated" : 0 } ] -},{ - "type" : "conv3d_transpose", - "comment" : "\nConvolution3D Transpose Operator.\n\nThe convolution transpose operation calculates the output based on the input, filter\nand dilations, strides, paddings, groups parameters. The size of each dimension of the\nparameters is checked in the infer-shape.\nInput(Input) and output(Output) are in NCDHW format. Where N is batch size, C is the\nnumber of channels, D is the depth of the feature, H is the height of the feature,\nand W is the width of the feature.\nFilter(Input) is in MCDHW format. Where M is the number of input feature channels,\nC is the number of output feature channels, D is the depth of the filter,H is the\nheight of the filter, and W is the width of the filter.\nParameters(strides, paddings) are three elements. These three elements represent\ndepth, height and width, respectively.\nThe input(X) size and output(Out) size may be different.\n\nExample: \n Input:\n Input shape: $(N, C_{in}, D_{in}, H_{in}, W_{in})$\n Filter shape: $(C_{in}, C_{out}, D_f, H_f, W_f)$\n Output:\n Output shape: $(N, C_{out}, D_{out}, H_{out}, W_{out})$\n Where\n $$\n D_{out} = (D_{in} - 1) * strides[0] - 2 * paddings[0] + D_f \\\\\n H_{out} = (H_{in} - 1) * strides[1] - 2 * paddings[1] + H_f \\\\\n W_{out} = (W_{in} - 1) * strides[2] - 2 * paddings[2] + W_f\n $$\n", - "inputs" : [ - { - "name" : "Input", - "comment" : "(Tensor) The input tensor of convolution transpose operator.The format of input tensor is NCDHW. Where N is batch size, C is the number of channels, D is the depth of the feature, H is the height of the feature, and W is the width of the feature.", - "duplicable" : 0, - "intermediate" : 0 - }, { - "name" : "Filter", - "comment" : "(Tensor) The filter tensor of convolution transpose operator.The format of the filter tensor is MCDHW, where M is the number of input feature channels, C is the number of output feature channels, D is the depth of the filter, H is the height of the filter, and W is the width of the filter.We enforce groups number == 1 and padding == 0 in the convolution3d transpose scenario.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "outputs" : [ - { - "name" : "Output", - "comment" : "(Tensor) The output tensor of convolution transpose operator.The format of output tensor is also NCDHW.Where N is batch size, C is the number of channels, D is the depth of the feature, H is the height of the feature, and W is the width of the feature.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "attrs" : [ - { - "name" : "dilations", - "type" : "int array", - "comment" : "(vector default:{1, 1, 1}), the dilations(d_dilation,h_dilation, w_dilation) of convolution transpose operator.", - "generated" : 0 - }, { - "name" : "strides", - "type" : "int array", - "comment" : "(vector default:{1, 1, 1}), the strides{d_stride, h_stride, w_stride} of convolution transpose operator.", - "generated" : 0 - }, { - "name" : "paddings", - "type" : "int array", - "comment" : "(vector default:{0, 0, 0}), paddings(d_pad, h_pad, w_pad) of convolution transpose operator.", - "generated" : 0 - } ] },{ "type" : "gru", "comment" : "\nGRU Operator implements part calculations of the complete GRU as following:\n\n\\f[\nupdate \\ gate: u_t = actGate(xu_t + W_u * h_{t-1} + b_u) \\\\\nreset \\ gate: r_t = actGate(xr_t + W_r * h_{t-1} + b_r) \\\\\noutput \\ candidate: {h}_t = actNode(xc_t + W_c * dot(r_t, h_{t-1}) + b_c) \\\\\noutput: h_t = dot((1 - u_t), h_{t-1}) + dot(u_t, {h}_t)\n\\f]\n\n@note To implement the complete GRU, fully-connected operator must be used \nbefore to feed xu, xr and xc as the Input of GRU operator.\n", @@ -618,136 +403,357 @@ "generated" : 0 } ] },{ - "type" : "recurrent", - "comment" : "\nStatic Length Recurrent Operator.\n\nThe static length recurrent operator can only operate on fixed size sequence\ndata, i.e. in each mini-batch, the sequence length of all inputs are the same.\n\n", + "type" : "warpctc", + "comment" : "\nAn operator integrating the open-source\n[warp-ctc](https://github.com/baidu-research/warp-ctc) library, which is used in\n[Deep Speech 2: End-toEnd Speech Recognition in English and Mandarin](\nhttps://arxiv.org/pdf/1512.02595v1.pdf),\nto compute Connectionist Temporal Classification (CTC) loss.\nIt can be aliased as softmax with ctc, since a native softmax activation is\ninterated to the warp-ctc library, to to normlize values for each row of the\ninput tensor.\n\nMore detail of CTC loss can be found by refering to\n[Connectionist Temporal Classification: Labelling Unsegmented Sequence Data with\nRecurrent Neural Networks](\nhttp://machinelearning.wustl.edu/mlpapers/paper_files/icml2006_GravesFGS06.pdf).\n", "inputs" : [ { - "name" : "inputs", - "comment" : "rnn inputs", - "duplicable" : 1, - "intermediate" : 0 - }, { - "name" : "initial_states", - "comment" : "rnn initial states", - "duplicable" : 1, + "name" : "Logits", + "comment" : "(LodTensor, default: LoDTensor), the unscaled probabilities of variable-length sequences, which is a 2-D Tensor with LoD information. It's shape is [Lp, num_classes + 1], where Lp is the sum of all input sequences' length and num_classes is the true number of classes (not including the blank label).", + "duplicable" : 0, "intermediate" : 0 }, { - "name" : "parameters", - "comment" : "Parameters are used by step block as its input. However, the input is not a sequence tensor. Every time step, each operator in step block just use the parameter directly.", - "duplicable" : 1, + "name" : "Label", + "comment" : "(LodTensor, default: LoDTensor), the ground truth of variable-length sequence, which is a 2-D Tensor with LoD information. It is of the shape [Lg, 1], where Lg is th sum of all labels' length.", + "duplicable" : 0, "intermediate" : 0 } ], "outputs" : [ { - "name" : "outputs", - "comment" : "The output sequence of RNN. The sequence length must be same.", - "duplicable" : 1, - "intermediate" : 0 + "name" : "WarpCTCGrad", + "comment" : "(Tensor, default: Tensor), a temporary output Tensor to store the gradients of warp-ctc, which is computed with loss together in one call. It is a 3-D Tensor of the shape [max_sequence_length, batch_size, num_classes + 1].", + "duplicable" : 0, + "intermediate" : 1 }, { - "name" : "step_scopes", - "comment" : "StepScopes contain all local variables in each time step.", + "name" : "Loss", + "comment" : "(Tensor, default: Tensor), the Connectionist Temporal Classification (CTC) loss, which is a 2-D Tensor of the shape [batch_size, 1]", "duplicable" : 0, "intermediate" : 0 } ], "attrs" : [ { - "name" : "ex_states", - "type" : "string array", - "comment" : "The ex-state variable names.\nThe ex-state means the state value in the ex-timestep or the previous time step\n[ex_states, states, initial_states@GRAD] must be the same order", + "name" : "blank", + "type" : "int", + "comment" : "(int, default: 0), the blank label of Connectionist Temporal Classification (CTC) loss, which is in the half-opened interval [0, num_classes + 1).", "generated" : 0 }, { - "name" : "states", - "type" : "string array", - "comment" : "The state variable names. [ex_states, states, initial_states@GRAD] must be the same order", + "name" : "norm_by_times", + "type" : "bool", + "comment" : "(bool, default: false), whether to normalize the gradients by the number of time-step, which is also the sequence's length.", "generated" : 0 + } ] +},{ + "type" : "cos_sim", + "comment" : "\nCosine Similarity Operator.\n\n$Out = X^T * Y / (\\sqrt{X^T * X} * \\sqrt{Y^T * Y})$\n\nThe input X and Y must have the same shape, except that the 1st dimension\nof input Y could be just 1 (different from input X), which will be\nbroadcasted to match the shape of input X before computing their cosine\nsimilarity.\n\nBoth the input X and Y can carry the LoD (Level of Details) information,\nor not. But the output only shares the LoD information with input X.\n\n", + "inputs" : [ + { + "name" : "X", + "comment" : "The 1st input of cos_sim op.", + "duplicable" : 0, + "intermediate" : 0 }, { - "name" : "sub_block", - "type" : "block id", - "comment" : "The step block inside RNN", + "name" : "Y", + "comment" : "The 2nd input of cos_sim op.", + "duplicable" : 0, + "intermediate" : 0 + } ], + "outputs" : [ + { + "name" : "Out", + "comment" : "The output of cos_sim op.", + "duplicable" : 0, + "intermediate" : 0 + }, { + "name" : "XNorm", + "comment" : "Norm of the first input, reduced along the 1st dimension.", + "duplicable" : 0, + "intermediate" : 1 + }, { + "name" : "YNorm", + "comment" : "Norm of the second input, reduced along the 1st dimension.", + "duplicable" : 0, + "intermediate" : 1 + } ], + "attrs" : [ ] +},{ + "type" : "conv3d", + "comment" : "\nConvolution3D Operator.\n\nThe convolution operation calculates the output based on the input, filter\nand strides, paddings, dilations, groups parameters. The size of each dimension of the\nparameters is checked in the infer-shape.\nInput(Input) and output(Output) are in NCDHW format, where N is batch\nsize, C is the number of channels,D is the depth of the feature, H is the height of\nthe feature, and W is the width of the feature.\nFilters(Input) is MCDHW format, where M is the number of output image channels,\nC is the number of input image channels, D is the depth of the filter,\nH is the height of the filter, and W is the width of the filter.\nParameters(strides, paddings, dilations) are three elements. These three elements\nrepresent depth, height and width, respectively.\nThe input(X) size and output(Out) size may be different.\n\nExample:\n Input:\n Input shape: $(N, C_{in}, D_{in}, H_{in}, W_{in})$\n Filter shape: $(C_{out}, C_{in}, D_f, H_f, W_f)$\n Output:\n Output shape: $(N, C_{out}, D_{out}, H_{out}, W_{out})$\n Where\n $$\n D_{out}= \\frac{(D_{in} + 2 * paddings[0] - (dilations[0] * (D_f - 1) + 1))}{ strides[0]}+ 1 \\\\\n H_{out}= \\frac{(H_{in} + 2 * paddings[1] - (dilations[1] * (H_f - 1) + 1))}{ strides[1]}+ 1 \\\\\n W_{out}= \\frac{(W_{in} + 2 * paddings[2] - (dilations[2] * (W_f - 1) + 1))}{ strides[2]}+ 1\n $$\n", + "inputs" : [ + { + "name" : "Input", + "comment" : "(Tensor) The input tensor of convolution operator. The format of input tensor is NCDHW. Where N is batch size, C is the number of channels, D is the depth of the feature, H is the height of the feature, and W is the width of the feature.", + "duplicable" : 0, + "intermediate" : 0 + }, { + "name" : "Filter", + "comment" : "(Tensor) The filter tensor of convolution operator. The format of the filter tensor is MCDHW, where M is the number of output image channels, C is the number of input image channels, D is the depth of the filter, H is the height of the filter, and W is the width of the filter.If the groups attribute is greater than 1, C equals the number of input image channels divided by the groups.", + "duplicable" : 0, + "intermediate" : 0 + } ], + "outputs" : [ + { + "name" : "Output", + "comment" : "(Tensor) The output tensor of convolution operator.The format of output tensor is also NCDHW.", + "duplicable" : 0, + "intermediate" : 0 + } ], + "attrs" : [ + { + "name" : "strides", + "type" : "int array", + "comment" : "(vector, default:{1, 1, 1}), the strides(d_stride, h_stride, w_stride) of convolution operator.", "generated" : 0 }, { - "name" : "reverse", - "type" : "bool", - "comment" : "Calculate RNN reversely or not.\nBy default reverse=False\n\nAssume the input data is [A, B, C, D]\n\nif reverse is False:\n the computation of RNN is like\n A B C D\n | | | |\n v v v v\n rnn -----> rnn -----> rnn ----> rnn\n | | | |\n v v v v\n o o o o\n\nif reverse is True\n the computation of RNN is like\n A B C D\n | | | |\n v v v v\n rnn <----- rnn <----- rnn <---- rnn\n | | | |\n v v v v\n o o o o\n", + "name" : "paddings", + "type" : "int array", + "comment" : "(vector, default:{0, 0, 0}), the paddings(d_pad, h_pad, w_pad) of convolution operator.", "generated" : 0 }, { - "name" : "is_train", + "name" : "groups", + "type" : "int", + "comment" : "(int default:1), the groups number of the convolution operator. According to grouped convolution in Alex Krizhevsky's Deep CNN paper: when group=2, the first half of the filters is only connected to the first half of the input channels, while the second half of the filters is only connected to the second half of the input channels.", + "generated" : 0 + }, { + "name" : "dilations", + "type" : "int array", + "comment" : "(vector default:{1, 1, 1}), the dilations(d_dilation, h_dilation, w_dilation) of convolution operator.", + "generated" : 0 + }, { + "name" : "use_cudnn", "type" : "bool", - "comment" : "", + "comment" : "(bool, default false) Only used in cudnn kernel, need install cudnn", + "generated" : 0 + }, { + "name" : "data_format", + "type" : "string", + "comment" : "(string, default NCHW) Only used in An optional string from: \"NHWC\", \"NCHW\". Defaults to \"NHWC\". Specify the data format of the output data, the input will be transformed automatically. ", + "generated" : 0 + }, { + "name" : "workspace_size_MB", + "type" : "int", + "comment" : "Only used in cudnn kernel. workspace size for cudnn, in MB, workspace is a section of GPU memory which will be allocated/freed each time the operator runs, larger workspace size can increase performance but also requires better hardware. This size should be chosen carefully.", "generated" : 0 } ] },{ - "type" : "warpctc", - "comment" : "\nAn operator integrating the open-source\n[warp-ctc](https://github.com/baidu-research/warp-ctc) library, which is used in\n[Deep Speech 2: End-toEnd Speech Recognition in English and Mandarin](\nhttps://arxiv.org/pdf/1512.02595v1.pdf),\nto compute Connectionist Temporal Classification (CTC) loss.\nIt can be aliased as softmax with ctc, since a native softmax activation is\ninterated to the warp-ctc library, to to normlize values for each row of the\ninput tensor.\n\nMore detail of CTC loss can be found by refering to\n[Connectionist Temporal Classification: Labelling Unsegmented Sequence Data with\nRecurrent Neural Networks](\nhttp://machinelearning.wustl.edu/mlpapers/paper_files/icml2006_GravesFGS06.pdf).\n", + "type" : "conv2d", + "comment" : "\nConvolution Operator.\n\nThe convolution operation calculates the output based on the input, filter\nand strides, paddings, dilations, groups parameters. The size of each dimension of the\nparameters is checked in the infer-shape.\nInput(Input) and Output(Output) are in NCHW format. Where N is batch\nsize, C is the number of channels, H is the height of the feature, and W is\nthe width of the feature.\nFilters(Input) is MCHW format. Where M is the number of output image channels, C is\nthe number of input image channels, H is the height of the filter, and W\nis the width of the filter.\nParameters(strides, paddings, dilations) are two elements. These two elements represent\nheight and width, respectively.\nThe input(X) size and output(Out) size may be different.\n\nExample:\n Input:\n Input shape: $(N, C_{in}, H_{in}, W_{in})$\n Filter shape: $(C_{out}, C_{in}, H_f, W_f)$\n Output:\n Output shape: $(N, C_{out}, H_{out}, W_{out})$\n Where\n$$\n H_{out}= \\frac{(H_{in} + 2 * paddings[0] - (dilations[0] * (H_f - 1) + 1))}{strides[0]}+ 1 \\\\\n W_{out}= \\frac{(W_{in} + 2 * paddings[1] - (dilations[1] * (W_f - 1) + 1))}{strides[1]}+ 1\n$$\n", "inputs" : [ { - "name" : "Logits", - "comment" : "(LodTensor, default: LoDTensor), the unscaled probabilities of variable-length sequences, which is a 2-D Tensor with LoD information. It's shape is [Lp, num_classes + 1], where Lp is the sum of all input sequences' length and num_classes is the true number of classes (not including the blank label).", + "name" : "Input", + "comment" : "(Tensor) The input tensor of convolution operator. The format of input tensor is NCHW, where N is batch size, C is the number of channels, H is the height of the feature, and W is the width of the feature.", "duplicable" : 0, "intermediate" : 0 }, { - "name" : "Label", - "comment" : "(LodTensor, default: LoDTensor), the ground truth of variable-length sequence, which is a 2-D Tensor with LoD information. It is of the shape [Lg, 1], where Lg is th sum of all labels' length.", + "name" : "Filter", + "comment" : "(Tensor) The filter tensor of convolution operator. The format of the filter tensor is MCHW, where M is the number of output image channels, C is the number of input image channels, H is the height of the filter, and W is the width of the filter. If the groups attribute is greater than 1, C equals the number of input image channels divided by the groups.", "duplicable" : 0, "intermediate" : 0 } ], "outputs" : [ { - "name" : "WarpCTCGrad", - "comment" : "(Tensor, default: Tensor), a temporary output Tensor to store the gradients of warp-ctc, which is computed with loss together in one call. It is a 3-D Tensor of the shape [max_sequence_length, batch_size, num_classes + 1].", - "duplicable" : 0, - "intermediate" : 1 - }, { - "name" : "Loss", - "comment" : "(Tensor, default: Tensor), the Connectionist Temporal Classification (CTC) loss, which is a 2-D Tensor of the shape [batch_size, 1]", + "name" : "Output", + "comment" : "(Tensor) The output tensor of convolution operator. The format of output tensor is also NCHW.", "duplicable" : 0, "intermediate" : 0 } ], "attrs" : [ { - "name" : "blank", + "name" : "strides", + "type" : "int array", + "comment" : "(vector default:{1, 1}), the strides(h_stride, w_stride) of convolution operator.", + "generated" : 0 + }, { + "name" : "paddings", + "type" : "int array", + "comment" : "(vector default:{0, 0}), the paddings(h_pad, w_pad) of convolution operator.", + "generated" : 0 + }, { + "name" : "groups", "type" : "int", - "comment" : "(int, default: 0), the blank label of Connectionist Temporal Classification (CTC) loss, which is in the half-opened interval [0, num_classes + 1).", + "comment" : "(int default:1), the groups number of the convolution operator. According to grouped convolution in Alex Krizhevsky's Deep CNN paper: when group=2, the first half of the filters is only connected to the first half of the input channels, while the second half of the filters is only connected to the second half of the input channels.", "generated" : 0 }, { - "name" : "norm_by_times", + "name" : "dilations", + "type" : "int array", + "comment" : "(vector default:{1, 1}), the dilations(h_dilation, w_dilation) of convolution operator.", + "generated" : 0 + }, { + "name" : "use_cudnn", "type" : "bool", - "comment" : "(bool, default: false), whether to normalize the gradients by the number of time-step, which is also the sequence's length.", + "comment" : "(bool, default false) Only used in cudnn kernel, need install cudnn", + "generated" : 0 + }, { + "name" : "data_format", + "type" : "string", + "comment" : "(string, default NCHW) Only used in An optional string from: \"NHWC\", \"NCHW\". Defaults to \"NHWC\". Specify the data format of the output data, the input will be transformed automatically. ", + "generated" : 0 + }, { + "name" : "workspace_size_MB", + "type" : "int", + "comment" : "Only used in cudnn kernel. Need set use_cudnn to true.workspace size for cudnn, in MB, workspace is a section of GPU memory which will be allocated/freed each time the operator runs, larger workspace size can increase performance but also requires better hardware. This size should be chosen carefully.", "generated" : 0 } ] },{ - "type" : "cos_sim", - "comment" : "\nCosine Similarity Operator.\n\n$Out = X^T * Y / (\\sqrt{X^T * X} * \\sqrt{Y^T * Y})$\n\nThe input X and Y must have the same shape, except that the 1st dimension\nof input Y could be just 1 (different from input X), which will be\nbroadcasted to match the shape of input X before computing their cosine\nsimilarity.\n\nBoth the input X and Y can carry the LoD (Level of Details) information,\nor not. But the output only shares the LoD information with input X.\n\n", + "type" : "pool3d", + "comment" : "\nPool3d Operator.\n\nThe pooling3d operation calculates the output based on\nthe input, pooling_type, ksize, strides, and paddings parameters.\nInput(X) and output(Out) are in NCDHW format, where N is batch\nsize, C is the number of channels, and D, H and W are the depth, height and\nwidth of the feature, respectively. Parameters(ksize, strides, paddings) \nare three elements. These three elements represent depth, height and \nwidth, respectively. The input(X) size and output(Out) size may be different.\n\nExample:\n Input:\n X shape: $(N, C, D_{in}, H_{in}, W_{in})$\n Output:\n Out shape: $(N, C, D_{out}, H_{out}, W_{out})$\n Where\n $$\n D_{out} = \\frac{(D_{in} - ksize[0] + 2 * paddings[0])}{strides[0]} + 1 \\\\\n H_{out} = \\frac{(H_{in} - ksize[1] + 2 * paddings[1])}{strides[1]} + 1 \\\\\n W_{out} = \\frac{(W_{in} - ksize[2] + 2 * paddings[2])}{strides[2]} + 1\n $$\n\n", "inputs" : [ { "name" : "X", - "comment" : "The 1st input of cos_sim op.", + "comment" : "(Tensor) The input tensor of pooling operator. The format of input tensor is NCDHW, where N is batch size, C is the number of channels, and D, H and W is the depth, height and width of the feature, respectively.", "duplicable" : 0, "intermediate" : 0 + } ], + "outputs" : [ + { + "name" : "Out", + "comment" : "(Tensor) The output tensor of pooling operator.The format of output tensor is also NCDHW, where N is batch size, C is the number of channels, and D, H and W is the depth, height and width of the feature, respectively.", + "duplicable" : 0, + "intermediate" : 0 + } ], + "attrs" : [ + { + "name" : "pooling_type", + "type" : "string", + "comment" : "(string) Pooling type, can be \"max\" for max-pooling and \"avg\" for average-pooling.", + "generated" : 0 }, { - "name" : "Y", - "comment" : "The 2nd input of cos_sim op.", + "name" : "ksize", + "type" : "int array", + "comment" : "(vector) The pooling window size(depth, height, width) of pooling operator. If global_pooling = true, ksize and paddings will be ignored.", + "generated" : 0 + }, { + "name" : "global_pooling", + "type" : "bool", + "comment" : "(bool, default false) Whether to use the global pooling. If global_pooling = true, ksize and paddings wille be ignored.", + "generated" : 0 + }, { + "name" : "strides", + "type" : "int array", + "comment" : "(vector, default {1,1,1}) Strides(depth, height, width) of the pooling operator.", + "generated" : 0 + }, { + "name" : "paddings", + "type" : "int array", + "comment" : "(vector, default {0,0,0}), paddings(depth, height, width) of pooling operator. If global_pooling = true, ksize and paddings will be ignored.", + "generated" : 0 + }, { + "name" : "use_cudnn", + "type" : "bool", + "comment" : "(bool, default false) Only used in cudnn kernel, need install cudnn", + "generated" : 0 + }, { + "name" : "data_format", + "type" : "string", + "comment" : "(string, default NCHW) Only used in An optional string from: \"NHWC\", \"NCHW\". Defaults to \"NHWC\". Specify the data format of the output data, the input will be transformed automatically. ", + "generated" : 0 + } ] +},{ + "type" : "pool2d", + "comment" : "\nPool2d Operator.\n\nThe pooling2d operation calculates the output based on\nthe input, pooling_type and ksize, strides, paddings parameters.\nInput(X) and output(Out) are in NCHW format, where N is batch size, C is the\nnumber of channels, H is the height of the feature, and W is the width of the feature.\nParameters(ksize, strides, paddings) are two elements.\nThese two elements represent height and width, respectively.\nThe input(X) size and output(Out) size may be different.\n\nExample: \n Input:\n X shape: $(N, C, H_{in}, W_{in})$\n Output:\n Out shape: $(N, C, H_{out}, W_{out})$\n Where\n $$ \n H_{out} = \\frac{(H_{in} - ksize[0] + 2 * paddings[0])}{strides[0]} + 1 \\\\\n W_{out} = \\frac{(W_{in} - ksize[1] + 2 * paddings[1])}{strides[1]} + 1\n $$\n\n", + "inputs" : [ + { + "name" : "X", + "comment" : "(Tensor) The input tensor of pooling operator. The format of input tensor is NCHW, where N is batch size, C is the number of channels, H is the height of the feature, and W is the width of the feature.", "duplicable" : 0, "intermediate" : 0 } ], "outputs" : [ { "name" : "Out", - "comment" : "The output of cos_sim op.", + "comment" : "(Tensor) The output tensor of pooling operator. The format of output tensor is also NCHW, where N is batch size, C is the number of channels, H is the height of the feature, and W is the width of the feature.", "duplicable" : 0, "intermediate" : 0 + } ], + "attrs" : [ + { + "name" : "pooling_type", + "type" : "string", + "comment" : "(string), pooling type, can be \"max\" for max-pooling and \"avg\" for average-pooling.", + "generated" : 0 }, { - "name" : "XNorm", - "comment" : "Norm of the first input, reduced along the 1st dimension.", + "name" : "ksize", + "type" : "int array", + "comment" : "(vector) The pooling window size(height, width) of the pooling operator. If global_pooling = true, ksize and paddings will be ignored.", + "generated" : 0 + }, { + "name" : "global_pooling", + "type" : "bool", + "comment" : "(bool, default false) Whether to use the global pooling. If global_pooling = true, ksize and paddings will be ignored.", + "generated" : 0 + }, { + "name" : "strides", + "type" : "int array", + "comment" : "(vector, default {1, 1}), strides(height, width) of pooling operator.", + "generated" : 0 + }, { + "name" : "paddings", + "type" : "int array", + "comment" : "(vector, default {0,0}), paddings(height, width) of pooling operator.If global_pooling = true, paddings and ksize will be ignored.", + "generated" : 0 + }, { + "name" : "use_cudnn", + "type" : "bool", + "comment" : "(bool, default false) Only used in cudnn kernel, need install cudnn", + "generated" : 0 + }, { + "name" : "data_format", + "type" : "string", + "comment" : "(string, default NCHW) Only used in An optional string from: \"NHWC\", \"NCHW\". Defaults to \"NHWC\". Specify the data format of the output data, the input will be transformed automatically. ", + "generated" : 0 + } ] +},{ + "type" : "conv3d_transpose", + "comment" : "\nConvolution3D Transpose Operator.\n\nThe convolution transpose operation calculates the output based on the input, filter\nand dilations, strides, paddings, groups parameters. The size of each dimension of the\nparameters is checked in the infer-shape.\nInput(Input) and output(Output) are in NCDHW format. Where N is batch size, C is the\nnumber of channels, D is the depth of the feature, H is the height of the feature,\nand W is the width of the feature.\nFilter(Input) is in MCDHW format. Where M is the number of input feature channels,\nC is the number of output feature channels, D is the depth of the filter,H is the\nheight of the filter, and W is the width of the filter.\nParameters(strides, paddings) are three elements. These three elements represent\ndepth, height and width, respectively.\nThe input(X) size and output(Out) size may be different.\n\nExample: \n Input:\n Input shape: $(N, C_{in}, D_{in}, H_{in}, W_{in})$\n Filter shape: $(C_{in}, C_{out}, D_f, H_f, W_f)$\n Output:\n Output shape: $(N, C_{out}, D_{out}, H_{out}, W_{out})$\n Where\n $$\n D_{out} = (D_{in} - 1) * strides[0] - 2 * paddings[0] + D_f \\\\\n H_{out} = (H_{in} - 1) * strides[1] - 2 * paddings[1] + H_f \\\\\n W_{out} = (W_{in} - 1) * strides[2] - 2 * paddings[2] + W_f\n $$\n", + "inputs" : [ + { + "name" : "Input", + "comment" : "(Tensor) The input tensor of convolution transpose operator.The format of input tensor is NCDHW. Where N is batch size, C is the number of channels, D is the depth of the feature, H is the height of the feature, and W is the width of the feature.", "duplicable" : 0, - "intermediate" : 1 + "intermediate" : 0 }, { - "name" : "YNorm", - "comment" : "Norm of the second input, reduced along the 1st dimension.", + "name" : "Filter", + "comment" : "(Tensor) The filter tensor of convolution transpose operator.The format of the filter tensor is MCDHW, where M is the number of input feature channels, C is the number of output feature channels, D is the depth of the filter, H is the height of the filter, and W is the width of the filter.We enforce groups number == 1 and padding == 0 in the convolution3d transpose scenario.", "duplicable" : 0, - "intermediate" : 1 + "intermediate" : 0 } ], - "attrs" : [ ] + "outputs" : [ + { + "name" : "Output", + "comment" : "(Tensor) The output tensor of convolution transpose operator.The format of output tensor is also NCDHW.Where N is batch size, C is the number of channels, D is the depth of the feature, H is the height of the feature, and W is the width of the feature.", + "duplicable" : 0, + "intermediate" : 0 + } ], + "attrs" : [ + { + "name" : "dilations", + "type" : "int array", + "comment" : "(vector default:{1, 1, 1}), the dilations(d_dilation,h_dilation, w_dilation) of convolution transpose operator.", + "generated" : 0 + }, { + "name" : "strides", + "type" : "int array", + "comment" : "(vector default:{1, 1, 1}), the strides{d_stride, h_stride, w_stride} of convolution transpose operator.", + "generated" : 0 + }, { + "name" : "paddings", + "type" : "int array", + "comment" : "(vector default:{0, 0, 0}), paddings(d_pad, h_pad, w_pad) of convolution transpose operator.", + "generated" : 0 + }, { + "name" : "use_cudnn", + "type" : "bool", + "comment" : "(bool, default false) Only used in cudnn kernel, need install cudnn", + "generated" : 0 + }, { + "name" : "data_format", + "type" : "string", + "comment" : "(string, default NCHW) Only used in An optional string from: \"NHWC\", \"NCHW\". Defaults to \"NHWC\". Specify the data format of the output data, the input will be transformed automatically. ", + "generated" : 0 + }, { + "name" : "workspace_size_MB", + "type" : "int", + "comment" : "Used in cudnn kernel only. workspace size for cudnn, in MB, workspace is a section of GPU memory which will be allocated/freed each time the operator runs, larger workspace size can increase performance but also requires better hardward. This size should be carefully setted.", + "generated" : 0 + } ] },{ "type" : "parallel_do", "comment" : "\nParallelDo Operator.\n", @@ -787,6 +793,65 @@ "comment" : "", "generated" : 0 } ] +},{ + "type" : "recurrent", + "comment" : "\nStatic Length Recurrent Operator.\n\nThe static length recurrent operator can only operate on fixed size sequence\ndata, i.e. in each mini-batch, the sequence length of all inputs are the same.\n\n", + "inputs" : [ + { + "name" : "inputs", + "comment" : "rnn inputs", + "duplicable" : 1, + "intermediate" : 0 + }, { + "name" : "initial_states", + "comment" : "rnn initial states", + "duplicable" : 1, + "intermediate" : 0 + }, { + "name" : "parameters", + "comment" : "Parameters are used by step block as its input. However, the input is not a sequence tensor. Every time step, each operator in step block just use the parameter directly.", + "duplicable" : 1, + "intermediate" : 0 + } ], + "outputs" : [ + { + "name" : "outputs", + "comment" : "The output sequence of RNN. The sequence length must be same.", + "duplicable" : 1, + "intermediate" : 0 + }, { + "name" : "step_scopes", + "comment" : "StepScopes contain all local variables in each time step.", + "duplicable" : 0, + "intermediate" : 0 + } ], + "attrs" : [ + { + "name" : "ex_states", + "type" : "string array", + "comment" : "The ex-state variable names.\nThe ex-state means the state value in the ex-timestep or the previous time step\n[ex_states, states, initial_states@GRAD] must be the same order", + "generated" : 0 + }, { + "name" : "states", + "type" : "string array", + "comment" : "The state variable names. [ex_states, states, initial_states@GRAD] must be the same order", + "generated" : 0 + }, { + "name" : "sub_block", + "type" : "block id", + "comment" : "The step block inside RNN", + "generated" : 0 + }, { + "name" : "reverse", + "type" : "bool", + "comment" : "Calculate RNN reversely or not.\nBy default reverse=False\n\nAssume the input data is [A, B, C, D]\n\nif reverse is False:\n the computation of RNN is like\n A B C D\n | | | |\n v v v v\n rnn -----> rnn -----> rnn ----> rnn\n | | | |\n v v v v\n o o o o\n\nif reverse is True\n the computation of RNN is like\n A B C D\n | | | |\n v v v v\n rnn <----- rnn <----- rnn <---- rnn\n | | | |\n v v v v\n o o o o\n", + "generated" : 0 + }, { + "name" : "is_train", + "type" : "bool", + "comment" : "", + "generated" : 0 + } ] },{ "type" : "save", "comment" : "\nSave operator\n\nThis operator will serialize and write a tensor variable to file on disk.\n", @@ -1093,91 +1158,55 @@ "generated" : 0 } ] },{ - "type" : "softsign", - "comment" : "\nSoftsign Activation Operator.\n\n$$out = \\frac{x}{1 + |x|}$$\n\n", - "inputs" : [ - { - "name" : "X", - "comment" : "Input of Softsign operator", - "duplicable" : 0, - "intermediate" : 0 - } ], - "outputs" : [ - { - "name" : "Out", - "comment" : "Output of Softsign operator", - "duplicable" : 0, - "intermediate" : 0 - } ], - "attrs" : [ ] -},{ - "type" : "square", - "comment" : "\nSquare Activation Operator.\n\n$out = x^2$\n\n", - "inputs" : [ - { - "name" : "X", - "comment" : "Input of Square operator", - "duplicable" : 0, - "intermediate" : 0 - } ], - "outputs" : [ - { - "name" : "Out", - "comment" : "Output of Square operator", - "duplicable" : 0, - "intermediate" : 0 - } ], - "attrs" : [ ] -},{ - "type" : "log", - "comment" : "\nLog Activation Operator.\n\n$out = \\ln(x)$\n\nNatural logarithm of x.\n\n", + "type" : "sqrt", + "comment" : "\nSqrt Activation Operator.\n\n$out = \\sqrt{x}$\n\n", "inputs" : [ { "name" : "X", - "comment" : "Input of Log operator", + "comment" : "Input of Sqrt operator", "duplicable" : 0, "intermediate" : 0 } ], "outputs" : [ { "name" : "Out", - "comment" : "Output of Log operator", + "comment" : "Output of Sqrt operator", "duplicable" : 0, "intermediate" : 0 } ], "attrs" : [ ] },{ - "type" : "reciprocal", - "comment" : "\nReciprocal Activation Operator.\n\n$$out = \\frac{1}{x}$$\n\n", + "type" : "softmax", + "comment" : "\nSoftmax Operator.\n\nThe input of the softmax operator is a 2-D tensor with shape N x K (N is the\nbatch_size, K is the dimension of input feature). The output tensor has the\nsame shape as the input tensor.\n\nFor each row of the input tensor, the softmax operator squashes the\nK-dimensional vector of arbitrary real values to a K-dimensional vector of real\nvalues in the range [0, 1] that add up to 1.\nIt computes the exponential of the given dimension and the sum of exponential\nvalues of all the other dimensions in the K-dimensional vector input.\nThen the ratio of the exponential of the given dimension and the sum of\nexponential values of all the other dimensions is the output of the softmax\noperator.\n\nFor each row $i$ and each column $j$ in Input(X), we have:\n $$Out[i, j] = \\frac{\\exp(X[i, j])}{\\sum_j(exp(X[i, j])}$$\n\n", "inputs" : [ { "name" : "X", - "comment" : "Input of Reciprocal operator", + "comment" : "The input tensor of softmax. 2-D with shape [batch_size, input_feature_dimensions].", "duplicable" : 0, "intermediate" : 0 } ], "outputs" : [ { "name" : "Out", - "comment" : "Output of Reciprocal operator", + "comment" : "The normalized values with the same shape as X.", "duplicable" : 0, "intermediate" : 0 } ], "attrs" : [ ] },{ - "type" : "softmax", - "comment" : "\nSoftmax Operator.\n\nThe input of the softmax operator is a 2-D tensor with shape N x K (N is the\nbatch_size, K is the dimension of input feature). The output tensor has the\nsame shape as the input tensor.\n\nFor each row of the input tensor, the softmax operator squashes the\nK-dimensional vector of arbitrary real values to a K-dimensional vector of real\nvalues in the range [0, 1] that add up to 1.\nIt computes the exponential of the given dimension and the sum of exponential\nvalues of all the other dimensions in the K-dimensional vector input.\nThen the ratio of the exponential of the given dimension and the sum of\nexponential values of all the other dimensions is the output of the softmax\noperator.\n\nFor each row $i$ and each column $j$ in Input(X), we have:\n $$Out[i, j] = \\frac{\\exp(X[i, j])}{\\sum_j(exp(X[i, j])}$$\n\n", + "type" : "lod_array_length", + "comment" : "\nLoDArrayLength Operator.\n\nThis operator obtains the length of lod tensor array:\n\n$$Out = len(X)$$\n\nNOTE: The output is a CPU Tensor since the control variable should be only in\nCPU and the length of LoDTensorArray should be used as control variables.\n\n", "inputs" : [ { "name" : "X", - "comment" : "The input tensor of softmax. 2-D with shape [batch_size, input_feature_dimensions].", + "comment" : "(LoDTensorArray) The input tensor array.", "duplicable" : 0, "intermediate" : 0 } ], "outputs" : [ { "name" : "Out", - "comment" : "The normalized values with the same shape as X.", + "comment" : "(Tensor) 1x1 CPU Tensor of length, int64_t", "duplicable" : 0, "intermediate" : 0 } ], @@ -1348,132 +1377,35 @@ } ], "outputs" : [ { - "name" : "Out", - "comment" : "The max sequence length.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "attrs" : [ ] -},{ - "type" : "multiplex", - "comment" : "\nMultiplex Operator.\n\nMultiplex multiple tensors according to the index provided by the index tensor.\n\nIds: the index tensor.\nX[0 : N - 1]: the candidate tensors for output (N >= 2).\nFor each index i from 0 to batchSize - 1, the output is the i-th row of the\nthe (Ids[i])-th tensor.\n\nFor i-th row of the output tensor:\n\n$$y[i] = x_{k}[i]$$\n\nwhere `y` is the output tensor, `x_{k}` is the k-th input tensor,\nand `k = Ids[i]`.\n\n", - "inputs" : [ - { - "name" : "Ids", - "comment" : "The index tensor of multiplex operator.", - "duplicable" : 0, - "intermediate" : 0 - }, { - "name" : "X", - "comment" : "The candidate tensors of multiplex operator.", - "duplicable" : 1, - "intermediate" : 0 - } ], - "outputs" : [ - { - "name" : "Out", - "comment" : "The output tensor of multiplex operator.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "attrs" : [ ] -},{ - "type" : "pool3d_cudnn", - "comment" : "\nPool3d Operator.\n\nThe pooling3d operation calculates the output based on\nthe input, pooling_type, ksize, strides, and paddings parameters.\nInput(X) and output(Out) are in NCDHW format, where N is batch\nsize, C is the number of channels, and D, H and W are the depth, height and\nwidth of the feature, respectively. Parameters(ksize, strides, paddings) \nare three elements. These three elements represent depth, height and \nwidth, respectively. The input(X) size and output(Out) size may be different.\n\nExample:\n Input:\n X shape: $(N, C, D_{in}, H_{in}, W_{in})$\n Output:\n Out shape: $(N, C, D_{out}, H_{out}, W_{out})$\n Where\n $$\n D_{out} = \\frac{(D_{in} - ksize[0] + 2 * paddings[0])}{strides[0]} + 1 \\\\\n H_{out} = \\frac{(H_{in} - ksize[1] + 2 * paddings[1])}{strides[1]} + 1 \\\\\n W_{out} = \\frac{(W_{in} - ksize[2] + 2 * paddings[2])}{strides[2]} + 1\n $$\n\n", - "inputs" : [ - { - "name" : "X", - "comment" : "(Tensor) The input tensor of pooling operator. The format of input tensor is NCDHW, where N is batch size, C is the number of channels, and D, H and W is the depth, height and width of the feature, respectively.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "outputs" : [ - { - "name" : "Out", - "comment" : "(Tensor) The output tensor of pooling operator.The format of output tensor is also NCDHW, where N is batch size, C is the number of channels, and D, H and W is the depth, height and width of the feature, respectively.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "attrs" : [ - { - "name" : "pooling_type", - "type" : "string", - "comment" : "(string) Pooling type, can be \"max\" for max-pooling and \"avg\" for average-pooling.", - "generated" : 0 - }, { - "name" : "ksize", - "type" : "int array", - "comment" : "(vector) The pooling window size(depth, height, width) of pooling operator. If global_pooling = true, ksize and paddings will be ignored.", - "generated" : 0 - }, { - "name" : "global_pooling", - "type" : "bool", - "comment" : "(bool, default false) Whether to use the global pooling. If global_pooling = true, ksize and paddings wille be ignored.", - "generated" : 0 - }, { - "name" : "strides", - "type" : "int array", - "comment" : "(vector, default {1,1,1}) Strides(depth, height, width) of the pooling operator.", - "generated" : 0 - }, { - "name" : "paddings", - "type" : "int array", - "comment" : "(vector, default {0,0,0}), paddings(depth, height, width) of pooling operator. If global_pooling = true, ksize and paddings will be ignored.", - "generated" : 0 - } ] -},{ - "type" : "pow", - "comment" : "\nPow Activation Operator.\n\n$out = x^{factor}$\n\n", - "inputs" : [ - { - "name" : "X", - "comment" : "Input of Pow operator", - "duplicable" : 0, - "intermediate" : 0 - } ], - "outputs" : [ - { - "name" : "Out", - "comment" : "Output of Pow operator", - "duplicable" : 0, - "intermediate" : 0 - } ], - "attrs" : [ - { - "name" : "factor", - "type" : "float", - "comment" : "The exponential factor of Pow", - "generated" : 0 - } ] + "name" : "Out", + "comment" : "The max sequence length.", + "duplicable" : 0, + "intermediate" : 0 + } ], + "attrs" : [ ] },{ - "type" : "lookup_table", - "comment" : "\nLookup Table Operator.\n\nThis operator is used to perform lookups on the parameter W,\nthen concatenated into a dense tensor.\n\nThe input Ids can carry the LoD (Level of Details) information,\nor not. And the output only shares the LoD information with input Ids.\n\n", + "type" : "multiplex", + "comment" : "\nMultiplex Operator.\n\nMultiplex multiple tensors according to the index provided by the index tensor.\n\nIds: the index tensor.\nX[0 : N - 1]: the candidate tensors for output (N >= 2).\nFor each index i from 0 to batchSize - 1, the output is the i-th row of the\nthe (Ids[i])-th tensor.\n\nFor i-th row of the output tensor:\n\n$$y[i] = x_{k}[i]$$\n\nwhere `y` is the output tensor, `x_{k}` is the k-th input tensor,\nand `k = Ids[i]`.\n\n", "inputs" : [ { - "name" : "W", - "comment" : "An input represents embedding tensors, which is a learnable parameter.", + "name" : "Ids", + "comment" : "The index tensor of multiplex operator.", "duplicable" : 0, "intermediate" : 0 }, { - "name" : "Ids", - "comment" : "An input with type int32 or int64 contains the ids to be looked up in W. Ids must be a column vector with rank = 2. The 2nd dimension size must be 1.", - "duplicable" : 0, + "name" : "X", + "comment" : "The candidate tensors of multiplex operator.", + "duplicable" : 1, "intermediate" : 0 } ], "outputs" : [ { "name" : "Out", - "comment" : "The lookup results, which have the same type as W.", + "comment" : "The output tensor of multiplex operator.", "duplicable" : 0, "intermediate" : 0 } ], - "attrs" : [ - { - "name" : "is_sparse", - "type" : "bool", - "comment" : "(boolean, default false) Sparse update", - "generated" : 0 - } ] + "attrs" : [ ] },{ "type" : "positive_negative_pair", "comment" : "\nPositiveNegativePairOp can be used to evaluate Learning To Rank(LTR) model's\nperformance.\n\nWithin some context, e.g. the \"query\", a LTR model generates scores for a list\nof items, which gives a partial order of the items. PositiveNegativePairOp\ntakes a list of reference rank order (Input(\"Label\")) and the model generated\nscores (Input(Score)) as inputs and counts the pairs that ranked correctly\nand incorrectly.\n", @@ -1649,6 +1581,24 @@ "comment" : "(float, default 0.0) L2 regularization strength.", "generated" : 0 } ] +},{ + "type" : "reciprocal", + "comment" : "\nReciprocal Activation Operator.\n\n$$out = \\frac{1}{x}$$\n\n", + "inputs" : [ + { + "name" : "X", + "comment" : "Input of Reciprocal operator", + "duplicable" : 0, + "intermediate" : 0 + } ], + "outputs" : [ + { + "name" : "Out", + "comment" : "Output of Reciprocal operator", + "duplicable" : 0, + "intermediate" : 0 + } ], + "attrs" : [ ] },{ "type" : "reduce_min", "comment" : "\n{ReduceOp} Operator.\n\nThis operator computes the min of input tensor along the given dimension. \nThe result tensor has 1 fewer dimension than the input unless keep_dim is true.\nIf reduce_all is true, just reduce along all dimensions and output a scalar.\n\n", @@ -1751,24 +1701,6 @@ "comment" : "(bool, default false) If true, output a scalar reduced along all dimensions.", "generated" : 0 } ] -},{ - "type" : "round", - "comment" : "\nRound Activation Operator.\n\n$out = [x]$\n\n", - "inputs" : [ - { - "name" : "X", - "comment" : "Input of Round operator", - "duplicable" : 0, - "intermediate" : 0 - } ], - "outputs" : [ - { - "name" : "Out", - "comment" : "Output of Round operator", - "duplicable" : 0, - "intermediate" : 0 - } ], - "attrs" : [ ] },{ "type" : "norm", "comment" : "\n \"Input shape: $(N, C, H, W)$\n Scale shape: $(C, 1)$\n Output shape: $(N, C, H, W)$\n Where\n forward\n $$\n [\\frac {x_{1}}{\\sqrt{\\sum{x_{i}^{2}}}} \\frac {x_{2}}{\\sqrt{\\sum{x_{i}^{2}}}} \\frac {x_{3}}{\\sqrt{\\sum{x_{i}^{2}}}} \\cdot \\cdot \\cdot \\frac {x_{n}}{\\sqrt{\\sum{x_{i}^{2}}}}]\n $$\n backward\n $$\n \\frac{\\frac{\\mathrm{d}L }{\\mathrm{d}y_{1}} - \\frac {x_{1}\\sum {\\frac{\\mathrm{d} L}{\\mathrm{d} y_{j}}}x_{j}}{\\sum x_{j}^{2}} }{\\sqrt{\\sum{x_{j}^{2}}}}\n $$\n ", @@ -1826,35 +1758,6 @@ "intermediate" : 0 } ], "attrs" : [ ] -},{ - "type" : "elementwise_sub", - "comment" : "\nLimited Elementwise Sub Operator.\n\nThe equation is:\n\n.. math::\n Out = X - Y\n\nX is a tensor of any dimension and the dimensions of tensor Y must be smaller than\nor equal to the dimensions of X. \n\nThere are two cases for this operator:\n1. The shape of Y is same with X;\n2. The shape of Y is a subset of X.\n\nFor case 2:\nY will be broadcasted to match the shape of X and axis should be \nthe starting dimension index for broadcasting Y onto X.\n\nFor example\n .. code-block:: python\n\n shape(X) = (2, 3, 4, 5), shape(Y) = (,)\n shape(X) = (2, 3, 4, 5), shape(Y) = (5,)\n shape(X) = (2, 3, 4, 5), shape(Y) = (4, 5)\n shape(X) = (2, 3, 4, 5), shape(Y) = (3, 4), with axis=1\n shape(X) = (2, 3, 4, 5), shape(Y) = (2), with axis=0\n\nEither of the inputs X and Y or none can carry the LoD (Level of Details) information. However, the output only shares the LoD information with input X.\n\n", - "inputs" : [ - { - "name" : "X", - "comment" : "(Tensor) The first input tensor of elementwise op", - "duplicable" : 0, - "intermediate" : 0 - }, { - "name" : "Y", - "comment" : "(Tensor) The second input tensor of elementwise op", - "duplicable" : 0, - "intermediate" : 0 - } ], - "outputs" : [ - { - "name" : "Out", - "comment" : "The output of elementwise op", - "duplicable" : 0, - "intermediate" : 0 - } ], - "attrs" : [ - { - "name" : "axis", - "type" : "int", - "comment" : "(int, default -1) The starting dimension index for broadcasting Y onto X", - "generated" : 0 - } ] },{ "type" : "rnn_memory_helper", "comment" : "", @@ -1975,55 +1878,6 @@ "intermediate" : 0 } ], "attrs" : [ ] -},{ - "type" : "conv3d_cudnn", - "comment" : "\nConvolution3D Operator.\n\nThe convolution operation calculates the output based on the input, filter\nand strides, paddings, dilations, groups parameters. The size of each dimension of the\nparameters is checked in the infer-shape.\nInput(Input) and output(Output) are in NCDHW format, where N is batch\nsize, C is the number of channels,D is the depth of the feature, H is the height of\nthe feature, and W is the width of the feature.\nFilters(Input) is MCDHW format, where M is the number of output image channels,\nC is the number of input image channels, D is the depth of the filter,\nH is the height of the filter, and W is the width of the filter.\nParameters(strides, paddings, dilations) are three elements. These three elements\nrepresent depth, height and width, respectively.\nThe input(X) size and output(Out) size may be different.\n\nExample:\n Input:\n Input shape: $(N, C_{in}, D_{in}, H_{in}, W_{in})$\n Filter shape: $(C_{out}, C_{in}, D_f, H_f, W_f)$\n Output:\n Output shape: $(N, C_{out}, D_{out}, H_{out}, W_{out})$\n Where\n $$\n D_{out}= \\frac{(D_{in} + 2 * paddings[0] - (dilations[0] * (D_f - 1) + 1))}{ strides[0]}+ 1 \\\\\n H_{out}= \\frac{(H_{in} + 2 * paddings[1] - (dilations[1] * (H_f - 1) + 1))}{ strides[1]}+ 1 \\\\\n W_{out}= \\frac{(W_{in} + 2 * paddings[2] - (dilations[2] * (W_f - 1) + 1))}{ strides[2]}+ 1\n $$\n", - "inputs" : [ - { - "name" : "Input", - "comment" : "(Tensor) The input tensor of convolution operator. The format of input tensor is NCDHW. Where N is batch size, C is the number of channels, D is the depth of the feature, H is the height of the feature, and W is the width of the feature.", - "duplicable" : 0, - "intermediate" : 0 - }, { - "name" : "Filter", - "comment" : "(Tensor) The filter tensor of convolution operator. The format of the filter tensor is MCDHW, where M is the number of output image channels, C is the number of input image channels, D is the depth of the filter, H is the height of the filter, and W is the width of the filter.If the groups attribute is greater than 1, C equals the number of input image channels divided by the groups.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "outputs" : [ - { - "name" : "Output", - "comment" : "(Tensor) The output tensor of convolution operator.The format of output tensor is also NCDHW.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "attrs" : [ - { - "name" : "strides", - "type" : "int array", - "comment" : "(vector, default:{1, 1, 1}), the strides(d_stride, h_stride, w_stride) of convolution operator.", - "generated" : 0 - }, { - "name" : "paddings", - "type" : "int array", - "comment" : "(vector, default:{0, 0, 0}), the paddings(d_pad, h_pad, w_pad) of convolution operator.", - "generated" : 0 - }, { - "name" : "groups", - "type" : "int", - "comment" : "(int default:1), the groups number of the convolution operator. According to grouped convolution in Alex Krizhevsky's Deep CNN paper: when group=2, the first half of the filters is only connected to the first half of the input channels, while the second half of the filters is only connected to the second half of the input channels.", - "generated" : 0 - }, { - "name" : "dilations", - "type" : "int array", - "comment" : "(vector default:{1, 1, 1}), the dilations(d_dilation, h_dilation, w_dilation) of convolution operator.", - "generated" : 0 - }, { - "name" : "workspace_size_MB", - "type" : "int", - "comment" : "workspace size for cudnn, in MB, workspace is a section of GPU memory which will be allocated/freed each time the operator runs, larger workspace size can increase performance but also requires better hardware. This size should be chosen carefully.", - "generated" : 0 - } ] },{ "type" : "sequence_expand", "comment" : "\nSequence Expand Operator.\n\nThis operator expands input(X) according to LOD of input(Y).\nFollowing are cases to better explain how this works:\nCase 1:\n\nGiven 2-level a LoDTensor input(X)\n X.lod = [[0, 2, 3],\n [0, 1, 3, 4]]\n X.data = [a, b, c, d]\n X.dims = [4, 1]\nand input(Y)\n Y.lod = [[0, 2, 4],\n [0, 3, 6, 7, 8]]\nwith condition len(Y.lod[-1]) -1 == X.dims[0]\nthen we get 2-level LoDTensor\n Out.lod = [[0, 2, 4],\n [0, 3, 6, 7, 8]]\n Out.data = [a, a, a, b, b, b, c, d]\n Out.dims = [8, 1]\n\nCase 2:\n\nGiven a 0-level LoDTensor input(X)\n X.data = [a, b, c]\n X.lod = NULL\n X.dims = [3, 1]\nand input(Y)\n Y.lod = [[0, 2, 3, 6]]\nwith condition len(Y.lod[-1]) -1 == X.dims[0]\nthen we get 1-level LoDTensor\n Out.lod = [[0, 2, 3, 6]]\n Out.data = [a, a, b, c, c, c]\n Out.dims = [6, 1]\n\nCase 3:\n\nGiven a 0-level LoDTensor input(X)\n X.data = [[a, b], [c, d], [e, f]]\n X.lod = NULL\n X.dims = [3, 2]\nand input(Y)\n Y.lod = [[0, 2, 3, 6]]\nwith condition len(Y.lod[-1]) -1 == X.dims[0]\nthen we get 1-level LoDTensor\n Out.lod = [[0, 2, 3, 6]]\n Out.data = [[a,b], [a,b] [c,d], [e, f], [e, f], [e, f]]\n Out.dims = [6, 2]\n\nCase 4:\n\nGiven 2-level a LoDTensor input(X)\n X.lod = [[0, 2, 3],\n [0, 1, 3, 4]]\n X.data = [a, b, c, d]\n X.dims = [4, 1]\nand input(Y)\n Y.lod = [[0, 2, 4],\n [0, 3, 6, 6, 8]]\nwith condition len(Y.lod[-1]) -1 == X.dims[0]\nthen we get 2-level LoDTensor\n Out.lod = [[0, 2, 4],\n [0, 3, 6, 6, 8]]\n Out.data = [a, a, a, b, b, b, d, d]\n Out.dims = [8, 1]\n\n\n", @@ -2381,6 +2235,59 @@ "comment" : "(float, default 1.0) The step size by which the input tensor will be incremented.", "generated" : 0 } ] +},{ + "type" : "log_loss", + "comment" : "\nLogLoss Operator.\n\nLog loss is a loss function used for binary classification. Log Loss quantifies\nthe accuracy of a classifier by penalising false classifications. Minimising the\nLog Loss is equivalent to maximising the accuracy of the classifier. We define\nPredicted as the values predicted by our model and Labels as the target ground\ntruth value. Log loss can evaluate how close the predicted values are to the\ntarget. The shapes of Predicted and Labels are both [batch_size, 1].\nThe equation is:\n\n$$\nLoss = - Labels * log(Predicted + \\epsilon) -\n (1 - Labels) * log(1 - Predicted + \\epsilon)\n$$\n\n", + "inputs" : [ + { + "name" : "Predicted", + "comment" : "The input value (Predicted) of Log loss op.Predicted is a 2-D tensor with shape [batch_size, 1].", + "duplicable" : 0, + "intermediate" : 0 + }, { + "name" : "Labels", + "comment" : "The target value (Labels) of Log loss op.Labels is a 2-D tensor with shape [batch_size, 1].", + "duplicable" : 0, + "intermediate" : 0 + } ], + "outputs" : [ + { + "name" : "Loss", + "comment" : "The output tensor with shape [batch_size, 1] which represents the log loss.", + "duplicable" : 0, + "intermediate" : 0 + } ], + "attrs" : [ + { + "name" : "epsilon", + "type" : "float", + "comment" : "Epsilon in log loss.", + "generated" : 0 + } ] +},{ + "type" : "pow", + "comment" : "\nPow Activation Operator.\n\n$out = x^{factor}$\n\n", + "inputs" : [ + { + "name" : "X", + "comment" : "Input of Pow operator", + "duplicable" : 0, + "intermediate" : 0 + } ], + "outputs" : [ + { + "name" : "Out", + "comment" : "Output of Pow operator", + "duplicable" : 0, + "intermediate" : 0 + } ], + "attrs" : [ + { + "name" : "factor", + "type" : "float", + "comment" : "The exponential factor of Pow", + "generated" : 0 + } ] },{ "type" : "unpool", "comment" : "\nInput shape is: $(N, C_{in}, H_{in}, W_{in})$, Output shape is:\n$(N, C_{out}, H_{out}, W_{out})$, where\n$$\nH_{out} = (H_{in}−1) * strides[0] − 2 * paddings[0] + ksize[0] \\\\\nW_{out} = (W_{in}−1) * strides[1] − 2 * paddings[1] + ksize[1]\n$$\nPaper: http://www.matthewzeiler.com/wp-content/uploads/2017/07/iccv2011.pdf\n", @@ -2454,63 +2361,121 @@ "comment" : "", "inputs" : [ { - "name" : "Out@GRAD", - "comment" : "", - "duplicable" : 0, - "intermediate" : 0 - }, { + "name" : "Out@GRAD", + "comment" : "", + "duplicable" : 0, + "intermediate" : 0 + }, { + "name" : "X", + "comment" : "", + "duplicable" : 0, + "intermediate" : 0 + }, { + "name" : "Out", + "comment" : "", + "duplicable" : 0, + "intermediate" : 0 + } ], + "outputs" : [ + { + "name" : "X@GRAD", + "comment" : "", + "duplicable" : 0, + "intermediate" : 0 + } ], + "attrs" : [ + { + "name" : "dtype", + "type" : "int", + "comment" : "(int, default 5 (FP32)) Output data type", + "generated" : 0 + } ] +},{ + "type" : "shrink_rnn_memory", + "comment" : "\nThis operator is used to shrink output batch of memory defined in dynamic RNN.\n\nDynamic RNN is able to handle variable-length sequences, in which, sequences in\na mini-batch are sorted by their lengths first. After that, the longest sequence\nbecomes the first one in the sorted batch, followed by the second longest, the\nthird longest, and so on. Dynamic RNN then slices a batch input timestep by\ntimestep from the sorted input. Once any sequence in the input batch reaches its\nend, memory defined in dynamicRNN has to shrink its outputs to adapt to the input\nbatch size for the next time step.\n", + "inputs" : [ + { + "name" : "X", + "comment" : "(LoDTensor) The RNN step memory to be shrinked.", + "duplicable" : 0, + "intermediate" : 0 + }, { + "name" : "RankTable", + "comment" : "(LoDRankTable) The lod_rank_table of dynamic RNN.", + "duplicable" : 0, + "intermediate" : 0 + }, { + "name" : "I", + "comment" : "(LoDTensor) The step index. The RNN step memory 'X' will be shrinked to match the size of the input of the index'th step.", + "duplicable" : 0, + "intermediate" : 0 + } ], + "outputs" : [ + { + "name" : "Out", + "comment" : "(LoDTensor) The shrinked RNN step memory.", + "duplicable" : 0, + "intermediate" : 0 + } ], + "attrs" : [ ] +},{ + "type" : "lod_reset", + "comment" : "LoDReset operator\n\nReset LoD of Input(X) into a new one specified by Input(TargetLoD) or\nAttr(target_lod), or set LoD for Input(X) if it doesn't have one.\nCurrently the lod_reset operator only supports the reset of level 0 LoD.\nAt least one of Input(TargetLoD) and Attr(target_lod) must be set,\nand if both of them are set, Input(TargetLoD) will be chosen as the\ntarget LoD.\n\nAn example:\nGiven a float LoDTensor X with shape (6, 1), its transpose form represents\n\n [1.0, 2.0, 3.0, 4.0, 5.0, 6.0],\n\nwith LoD = [[0, 2, 5, 6]] and the three (transposed) sequences look like\n\n [1.0, 2.0], [3.0, 4.0, 5.0], [6.0].\n\nIf target LoD = [0, 4, 6], the lod_reset operator will reset the LoD and\nthe sequences that the LoDTensor Output(Out) contains becomes:\n\n [1.0, 2.0, 3.0, 4.0], [5.0, 6.0].\n\n", + "inputs" : [ + { "name" : "X", - "comment" : "", + "comment" : "(LoDTensor) The input tensor of lod_reset operator.", "duplicable" : 0, "intermediate" : 0 }, { - "name" : "Out", - "comment" : "", + "name" : "TargetLoD", + "comment" : "(Tensor, optional) The target level 0 LoD from Input().", "duplicable" : 0, "intermediate" : 0 } ], "outputs" : [ { - "name" : "X@GRAD", - "comment" : "", + "name" : "Out", + "comment" : "(LoDTensor) The output tensor of lod_reset operator.", "duplicable" : 0, "intermediate" : 0 } ], "attrs" : [ { - "name" : "dtype", - "type" : "int", - "comment" : "(int, default 5 (FP32)) Output data type", + "name" : "target_lod", + "type" : "int array", + "comment" : "The target level 0 LoD from Attr().", "generated" : 0 } ] },{ - "type" : "shrink_rnn_memory", - "comment" : "\nThis operator is used to shrink output batch of memory defined in dynamic RNN.\n\nDynamic RNN is able to handle variable-length sequences, in which, sequences in\na mini-batch are sorted by their lengths first. After that, the longest sequence\nbecomes the first one in the sorted batch, followed by the second longest, the\nthird longest, and so on. Dynamic RNN then slices a batch input timestep by\ntimestep from the sorted input. Once any sequence in the input batch reaches its\nend, memory defined in dynamicRNN has to shrink its outputs to adapt to the input\nbatch size for the next time step.\n", + "type" : "elementwise_sub", + "comment" : "\nLimited Elementwise Sub Operator.\n\nThe equation is:\n\n.. math::\n Out = X - Y\n\nX is a tensor of any dimension and the dimensions of tensor Y must be smaller than\nor equal to the dimensions of X. \n\nThere are two cases for this operator:\n1. The shape of Y is same with X;\n2. The shape of Y is a subset of X.\n\nFor case 2:\nY will be broadcasted to match the shape of X and axis should be \nthe starting dimension index for broadcasting Y onto X.\n\nFor example\n .. code-block:: python\n\n shape(X) = (2, 3, 4, 5), shape(Y) = (,)\n shape(X) = (2, 3, 4, 5), shape(Y) = (5,)\n shape(X) = (2, 3, 4, 5), shape(Y) = (4, 5)\n shape(X) = (2, 3, 4, 5), shape(Y) = (3, 4), with axis=1\n shape(X) = (2, 3, 4, 5), shape(Y) = (2), with axis=0\n\nEither of the inputs X and Y or none can carry the LoD (Level of Details) information. However, the output only shares the LoD information with input X.\n\n", "inputs" : [ { "name" : "X", - "comment" : "(LoDTensor) The RNN step memory to be shrinked.", - "duplicable" : 0, - "intermediate" : 0 - }, { - "name" : "RankTable", - "comment" : "(LoDRankTable) The lod_rank_table of dynamic RNN.", + "comment" : "(Tensor) The first input tensor of elementwise op", "duplicable" : 0, "intermediate" : 0 }, { - "name" : "I", - "comment" : "(LoDTensor) The step index. The RNN step memory 'X' will be shrinked to match the size of the input of the index'th step.", + "name" : "Y", + "comment" : "(Tensor) The second input tensor of elementwise op", "duplicable" : 0, "intermediate" : 0 } ], "outputs" : [ { "name" : "Out", - "comment" : "(LoDTensor) The shrinked RNN step memory.", + "comment" : "The output of elementwise op", "duplicable" : 0, "intermediate" : 0 } ], - "attrs" : [ ] + "attrs" : [ + { + "name" : "axis", + "type" : "int", + "comment" : "(int, default -1) The starting dimension index for broadcasting Y onto X", + "generated" : 0 + } ] },{ "type" : "logical_and", "comment" : "logical_and Operator\n\nIt operates element-wise on X and Y, and returns the Out. X, Y and Out are N-dim boolean tensors.\nEach element of Out is calculated by $$Out = X \\&\\& Y$$\n", @@ -2571,53 +2536,30 @@ } ], "attrs" : [ ] },{ - "type" : "softplus", - "comment" : "\nSoftplus Activation Operator.\n\n$out = \\ln(1 + e^{x})$\n\n", + "type" : "square", + "comment" : "\nSquare Activation Operator.\n\n$out = x^2$\n\n", "inputs" : [ { "name" : "X", - "comment" : "Input of Softplus operator", + "comment" : "Input of Square operator", "duplicable" : 0, "intermediate" : 0 } ], "outputs" : [ { "name" : "Out", - "comment" : "Output of Softplus operator", + "comment" : "Output of Square operator", "duplicable" : 0, "intermediate" : 0 } ], "attrs" : [ ] },{ - "type" : "get_places", - "comment" : "\nReturns a list of places based on flags. The list will be used for parallel\nexecution.\n", - "inputs" : [ ], - "outputs" : [ - { - "name" : "Out", - "comment" : "vector of Place", - "duplicable" : 0, - "intermediate" : 0 - } ], - "attrs" : [ - { - "name" : "device_count", - "type" : "int", - "comment" : "device count", - "generated" : 0 - }, { - "name" : "device_type", - "type" : "string", - "comment" : "device type", - "generated" : 0 - } ] -},{ - "type" : "read_from_array", - "comment" : "\nReadFromArray Operator.\n\nRead a LoDTensor from a LoDTensor Array.\n\nAssume $T$ is LoDTensor, $i$ is the subscript of the array, and $A$ is the array. The\nequation is\n\n$$T = A[i]$$\n\n", + "type" : "write_to_array", + "comment" : "\nWriteToArray Operator.\n\nThis operator writes a LoDTensor to a LoDTensor array.\n\nAssume $T$ is LoDTensor, $i$ is the subscript of the array, and $A$ is the array. The\nequation is\n\n$$A[i] = T$$\n\n", "inputs" : [ { "name" : "X", - "comment" : "(TensorArray) the array will be read from.", + "comment" : "(LoDTensor) the tensor will be written to tensor array", "duplicable" : 0, "intermediate" : 0 }, { @@ -2629,7 +2571,7 @@ "outputs" : [ { "name" : "Out", - "comment" : "(LoDTensor) the tensor will be read from.", + "comment" : "(TensorArray) the tensor array will be written", "duplicable" : 0, "intermediate" : 0 } ], @@ -2688,50 +2630,6 @@ "comment" : "(int) Number of classes to be evaluated.", "generated" : 0 } ] -},{ - "type" : "conv2d_transpose_cudnn", - "comment" : "\nConvolution2D Transpose Operator.\n\nThe convolution transpose operation calculates the output based on the input, filter\nand dilations, strides, paddings, groups parameters. The size of each dimension of the\nparameters is checked in the infer-shape.\nInput(Input) and output(Output) are in NCHW format. Where N is batchsize, C is the\nnumber of channels, H is the height of the feature, and W is the width of the feature.\nFilter(Input) is in MCHW format. Where M is the number of input feature channels,\nC is the number of output feature channels, H is the height of the filter,\nand W is the width of the filter.\nParameters(strides, paddings) are two elements. These two elements represent height\nand width, respectively.\nThe input(X) size and output(Out) size may be different.\n\nExample:\n Input:\n Input shape: $(N, C_{in}, H_{in}, W_{in})$\n Filter shape: $(C_{in}, C_{out}, H_f, W_f)$\n Output:\n Output shape: $(N, C_{out}, H_{out}, W_{out})$\n Where\n $$\n H_{out} = (H_{in} - 1) * strides[0] - 2 * paddings[0] + H_f \\\\\n W_{out} = (W_{in} - 1) * strides[1] - 2 * paddings[1] + W_f\n $$\n", - "inputs" : [ - { - "name" : "Input", - "comment" : "(Tensor) The input tensor of convolution transpose operator. The format of input tensor is NCHW. Where N is batch size, C is the number of input channels, H is the height of the feature, and W is the width of the feature.", - "duplicable" : 0, - "intermediate" : 0 - }, { - "name" : "Filter", - "comment" : "(Tensor) The filter tensor of convolution transpose operator. The format of the filter tensor is MCHW, where M is the number of input feature channels, C is the number of output feature channels,H is the height of the filter, and W is the width of the filter. We enforce groups number == 1 in the convolution transpose scenario.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "outputs" : [ - { - "name" : "Output", - "comment" : "(Tensor) The output tensor of convolution transpose operator. The format of output tensor is also NCHW.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "attrs" : [ - { - "name" : "dilations", - "type" : "int array", - "comment" : "(vector default:{1, 1}), the dilations(h_dilation, w_dilation) of convolution transpose operator.", - "generated" : 0 - }, { - "name" : "strides", - "type" : "int array", - "comment" : "(vector default:{1, 1}), the strides(h_stride, w_stride) of convolution transpose operator.", - "generated" : 0 - }, { - "name" : "paddings", - "type" : "int array", - "comment" : "(vector default:{0, 0}), the paddings(h_pad, w_pad) of convolution transpose operator.", - "generated" : 0 - }, { - "name" : "workspace_size_MB", - "type" : "int", - "comment" : "workspace size for cudnn, in MB, workspace is a section of GPU memory which will be allocated/freed each time the operator runs, larger workspace size can increase performance but also requires better hardward. This size should be carefully setted.", - "generated" : 0 - } ] },{ "type" : "merge_lod_tensor", "comment" : "\n Merge True and False branches of LoDTensor into a single Output,\n with a mask at certain lod level. X is used to obtain complete\n lod information. Please refer to SplitLoDTensorOp.", @@ -3012,6 +2910,104 @@ "comment" : "(bool, default false) If true, output a scalar reduced along all dimensions.", "generated" : 0 } ] +},{ + "type" : "stanh", + "comment" : "\nSTanh Activation Operator.\n\n$$out = b * \\frac{e^{a * x} - e^{-a * x}}{e^{a * x} + e^{-a * x}}$$\n\n", + "inputs" : [ + { + "name" : "X", + "comment" : "Input of STanh operator", + "duplicable" : 0, + "intermediate" : 0 + } ], + "outputs" : [ + { + "name" : "Out", + "comment" : "Output of STanh operator", + "duplicable" : 0, + "intermediate" : 0 + } ], + "attrs" : [ + { + "name" : "scale_a", + "type" : "float", + "comment" : "The scale parameter of a for the input", + "generated" : 0 + }, { + "name" : "scale_b", + "type" : "float", + "comment" : "The scale parameter of b for the input", + "generated" : 0 + } ] +},{ + "type" : "adamax", + "comment" : "\nAdamax Optimizer.\n\nWe implement the Adamax optimizer from Section 7 of the Adam\npaper: https://arxiv.org/abs/1412.6980. Adamax is a variant of the\nAdam algorithm based on the infinity norm.\n\nAdamax updates:\n\n$$\nmoment\\_out = \\beta_1 * moment + (1 - \\beta_1) * grad \\\\\ninf\\_norm\\_out = max(\\beta_2 * inf\\_norm + \\epsilon, |grad|) \\\\\nlearning\\_rate = \\frac{learning\\_rate}{1 - \\beta_{1\\_pow}} \\\\\nparam\\_out = param - learning\\_rate * \\frac{moment\\_out}{inf\\_norm\\_out}\n$$\n\nThe original paper does not have an epsilon attribute.\nHowever, it is added here for numerical stability to prevent the\ndivision by 0 error.\n\n", + "inputs" : [ + { + "name" : "Param", + "comment" : "(Tensor) Input parameter", + "duplicable" : 0, + "intermediate" : 0 + }, { + "name" : "Grad", + "comment" : "(Tensor) Input gradient", + "duplicable" : 0, + "intermediate" : 0 + }, { + "name" : "LearningRate", + "comment" : "(Tensor) Learning rate", + "duplicable" : 0, + "intermediate" : 0 + }, { + "name" : "Moment", + "comment" : "(Tensor) First moment", + "duplicable" : 0, + "intermediate" : 0 + }, { + "name" : "InfNorm", + "comment" : "(Tensor) Input exponentially weighted infinity norm", + "duplicable" : 0, + "intermediate" : 0 + }, { + "name" : "Beta1Pow", + "comment" : "(Tensor) Input beta1 power accumulator", + "duplicable" : 0, + "intermediate" : 0 + } ], + "outputs" : [ + { + "name" : "ParamOut", + "comment" : "(Tensor) Output parameter", + "duplicable" : 0, + "intermediate" : 0 + }, { + "name" : "MomentOut", + "comment" : "(Tensor) Output first moment", + "duplicable" : 0, + "intermediate" : 0 + }, { + "name" : "InfNormOut", + "comment" : "(Tensor) Output exponentially weighted infinity norm", + "duplicable" : 0, + "intermediate" : 0 + } ], + "attrs" : [ + { + "name" : "beta1", + "type" : "float", + "comment" : "(float, default 0.9) Exponential decay rate for the 1st moment estimates.", + "generated" : 0 + }, { + "name" : "beta2", + "type" : "float", + "comment" : "(float, default 0.999) exponential decay rate for the weighted infinity norm estimates.", + "generated" : 0 + }, { + "name" : "epsilon", + "type" : "float", + "comment" : "(float, default 1.0e-8) Constant for numerical stability", + "generated" : 0 + } ] },{ "type" : "tanh_shrink", "comment" : "\nTanhShrink Activation Operator.\n\n$$out = x - \\frac{e^{x} - e^{-x}}{e^{x} + e^{-x}}$$\n\n", @@ -3049,47 +3045,32 @@ } ], "attrs" : [ ] },{ - "type" : "pool2d_cudnn", - "comment" : "\nPool2d Operator.\n\nThe pooling2d operation calculates the output based on\nthe input, pooling_type and ksize, strides, paddings parameters.\nInput(X) and output(Out) are in NCHW format, where N is batch size, C is the\nnumber of channels, H is the height of the feature, and W is the width of the feature.\nParameters(ksize, strides, paddings) are two elements.\nThese two elements represent height and width, respectively.\nThe input(X) size and output(Out) size may be different.\n\nExample: \n Input:\n X shape: $(N, C, H_{in}, W_{in})$\n Output:\n Out shape: $(N, C, H_{out}, W_{out})$\n Where\n $$ \n H_{out} = \\frac{(H_{in} - ksize[0] + 2 * paddings[0])}{strides[0]} + 1 \\\\\n W_{out} = \\frac{(W_{in} - ksize[1] + 2 * paddings[1])}{strides[1]} + 1\n $$\n\n", + "type" : "lookup_table", + "comment" : "\nLookup Table Operator.\n\nThis operator is used to perform lookups on the parameter W,\nthen concatenated into a dense tensor.\n\nThe input Ids can carry the LoD (Level of Details) information,\nor not. And the output only shares the LoD information with input Ids.\n\n", "inputs" : [ { - "name" : "X", - "comment" : "(Tensor) The input tensor of pooling operator. The format of input tensor is NCHW, where N is batch size, C is the number of channels, H is the height of the feature, and W is the width of the feature.", + "name" : "W", + "comment" : "An input represents embedding tensors, which is a learnable parameter.", "duplicable" : 0, "intermediate" : 0 - } ], - "outputs" : [ - { - "name" : "Out", - "comment" : "(Tensor) The output tensor of pooling operator. The format of output tensor is also NCHW, where N is batch size, C is the number of channels, H is the height of the feature, and W is the width of the feature.", + }, { + "name" : "Ids", + "comment" : "An input with type int32 or int64 contains the ids to be looked up in W. Ids must be a column vector with rank = 2. The 2nd dimension size must be 1.", "duplicable" : 0, "intermediate" : 0 } ], - "attrs" : [ + "outputs" : [ { - "name" : "pooling_type", - "type" : "string", - "comment" : "(string), pooling type, can be \"max\" for max-pooling and \"avg\" for average-pooling.", - "generated" : 0 - }, { - "name" : "ksize", - "type" : "int array", - "comment" : "(vector) The pooling window size(height, width) of the pooling operator. If global_pooling = true, ksize and paddings will be ignored.", - "generated" : 0 - }, { - "name" : "global_pooling", + "name" : "Out", + "comment" : "The lookup results, which have the same type as W.", + "duplicable" : 0, + "intermediate" : 0 + } ], + "attrs" : [ + { + "name" : "is_sparse", "type" : "bool", - "comment" : "(bool, default false) Whether to use the global pooling. If global_pooling = true, ksize and paddings will be ignored.", - "generated" : 0 - }, { - "name" : "strides", - "type" : "int array", - "comment" : "(vector, default {1, 1}), strides(height, width) of pooling operator.", - "generated" : 0 - }, { - "name" : "paddings", - "type" : "int array", - "comment" : "(vector, default {0,0}), paddings(height, width) of pooling operator.If global_pooling = true, paddings and ksize will be ignored.", + "comment" : "(boolean, default false) Sparse update", "generated" : 0 } ] },{ @@ -3176,6 +3157,21 @@ "type" : "int array", "comment" : "(vector default:{0, 0}), the paddings(h_pad, w_pad) of convolution transpose operator.", "generated" : 0 + }, { + "name" : "use_cudnn", + "type" : "bool", + "comment" : "(bool, default false) Only used in cudnn kernel, need install cudnn", + "generated" : 0 + }, { + "name" : "data_format", + "type" : "string", + "comment" : "(string, default NCHW) Only used in An optional string from: \"NHWC\", \"NCHW\". Defaults to \"NHWC\". Specify the data format of the output data, the input will be transformed automatically. ", + "generated" : 0 + }, { + "name" : "workspace_size_MB", + "type" : "int", + "comment" : "Used in cudnn kernel only. workspace size for cudnn, in MB, workspace is a section of GPU memory which will be allocated/freed each time the operator runs, larger workspace size can increase performance but also requires better hardward. This size should be carefully setted.", + "generated" : 0 } ] },{ "type" : "less_than", @@ -3278,268 +3274,53 @@ "comment" : "(vector) Target shape of reshape operator.", "generated" : 0 } ] -},{ - "type" : "log_loss", - "comment" : "\nLogLoss Operator.\n\nLog loss is a loss function used for binary classification. Log Loss quantifies\nthe accuracy of a classifier by penalising false classifications. Minimising the\nLog Loss is equivalent to maximising the accuracy of the classifier. We define\nPredicted as the values predicted by our model and Labels as the target ground\ntruth value. Log loss can evaluate how close the predicted values are to the\ntarget. The shapes of Predicted and Labels are both [batch_size, 1].\nThe equation is:\n\n$$\nLoss = - Labels * log(Predicted + \\epsilon) -\n (1 - Labels) * log(1 - Predicted + \\epsilon)\n$$\n\n", - "inputs" : [ - { - "name" : "Predicted", - "comment" : "The input value (Predicted) of Log loss op.Predicted is a 2-D tensor with shape [batch_size, 1].", - "duplicable" : 0, - "intermediate" : 0 - }, { - "name" : "Labels", - "comment" : "The target value (Labels) of Log loss op.Labels is a 2-D tensor with shape [batch_size, 1].", - "duplicable" : 0, - "intermediate" : 0 - } ], - "outputs" : [ - { - "name" : "Loss", - "comment" : "The output tensor with shape [batch_size, 1] which represents the log loss.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "attrs" : [ - { - "name" : "epsilon", - "type" : "float", - "comment" : "Epsilon in log loss.", - "generated" : 0 - } ] -},{ - "type" : "sqrt", - "comment" : "\nSqrt Activation Operator.\n\n$out = \\sqrt{x}$\n\n", - "inputs" : [ - { - "name" : "X", - "comment" : "Input of Sqrt operator", - "duplicable" : 0, - "intermediate" : 0 - } ], - "outputs" : [ - { - "name" : "Out", - "comment" : "Output of Sqrt operator", - "duplicable" : 0, - "intermediate" : 0 - } ], - "attrs" : [ ] -},{ - "type" : "lod_reset", - "comment" : "LoDReset operator\n\nReset LoD of Input(X) into a new one specified by Input(TargetLoD) or\nAttr(target_lod), or set LoD for Input(X) if it doesn't have one.\nCurrently the lod_reset operator only supports the reset of level 0 LoD.\nAt least one of Input(TargetLoD) and Attr(target_lod) must be set,\nand if both of them are set, Input(TargetLoD) will be chosen as the\ntarget LoD.\n\nAn example:\nGiven a float LoDTensor X with shape (6, 1), its transpose form represents\n\n [1.0, 2.0, 3.0, 4.0, 5.0, 6.0],\n\nwith LoD = [[0, 2, 5, 6]] and the three (transposed) sequences look like\n\n [1.0, 2.0], [3.0, 4.0, 5.0], [6.0].\n\nIf target LoD = [0, 4, 6], the lod_reset operator will reset the LoD and\nthe sequences that the LoDTensor Output(Out) contains becomes:\n\n [1.0, 2.0, 3.0, 4.0], [5.0, 6.0].\n\n", - "inputs" : [ - { - "name" : "X", - "comment" : "(LoDTensor) The input tensor of lod_reset operator.", - "duplicable" : 0, - "intermediate" : 0 - }, { - "name" : "TargetLoD", - "comment" : "(Tensor, optional) The target level 0 LoD from Input().", - "duplicable" : 0, - "intermediate" : 0 - } ], - "outputs" : [ - { - "name" : "Out", - "comment" : "(LoDTensor) The output tensor of lod_reset operator.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "attrs" : [ - { - "name" : "target_lod", - "type" : "int array", - "comment" : "The target level 0 LoD from Attr().", - "generated" : 0 - } ] -},{ - "type" : "write_to_array", - "comment" : "\nWriteToArray Operator.\n\nThis operator writes a LoDTensor to a LoDTensor array.\n\nAssume $T$ is LoDTensor, $i$ is the subscript of the array, and $A$ is the array. The\nequation is\n\n$$A[i] = T$$\n\n", - "inputs" : [ - { - "name" : "X", - "comment" : "(LoDTensor) the tensor will be written to tensor array", - "duplicable" : 0, - "intermediate" : 0 - }, { - "name" : "I", - "comment" : "(Tensor) the subscript index in tensor array. The number of element should be 1", - "duplicable" : 0, - "intermediate" : 0 - } ], - "outputs" : [ - { - "name" : "Out", - "comment" : "(TensorArray) the tensor array will be written", - "duplicable" : 0, - "intermediate" : 0 - } ], - "attrs" : [ ] -},{ - "type" : "lod_array_length", - "comment" : "\nLoDArrayLength Operator.\n\nThis operator obtains the length of lod tensor array:\n\n$$Out = len(X)$$\n\nNOTE: The output is a CPU Tensor since the control variable should be only in\nCPU and the length of LoDTensorArray should be used as control variables.\n\n", - "inputs" : [ - { - "name" : "X", - "comment" : "(LoDTensorArray) The input tensor array.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "outputs" : [ - { - "name" : "Out", - "comment" : "(Tensor) 1x1 CPU Tensor of length, int64_t", - "duplicable" : 0, - "intermediate" : 0 - } ], - "attrs" : [ ] },{ "type" : "edit_distance", "comment" : "\n\nEditDistance operator computes the edit distances between a batch of hypothesis\nstrings and their references.\n\nEdit distance, also called Levenshtein distance, measures how dissimilar two strings \nare by counting the minimum number of operations to transform one string into anthor. \nHere the operations include insertion, deletion, and substitution. For example, \ngiven hypothesis string A = \"kitten\" and reference B = \"sitting\", the edit distance \nis 3 for A will be transformed into B at least after two substitutions and one \ninsertion:\n \n \"kitten\" -> \"sitten\" -> \"sittin\" -> \"sitting\"\n\nInput(Hyps) is a LoDTensor consisting of all the hypothesis strings with the total \nnumber denoted by `batch_size`, and the separation is specified by the LoD information. \nAnd the `batch_size` reference strings are arranged in order in the same way in the \nLoDTensor Input(Refs).\n\nOutput(Out) contains the `batch_size` results and each stands for the edit stance \nfor a pair of strings respectively. If Attr(normalized) is true, the edit distance \nwill be divided by the length of reference string.\n", "inputs" : [ - { - "name" : "Hyps", - "comment" : "(2-D LoDTensor, 2nd dim. equal to 1) The indices for hypothesis strings.", - "duplicable" : 0, - "intermediate" : 0 - }, { - "name" : "Refs", - "comment" : "(2-D LoDTensor, 2nd dim. equal to 1) The indices for reference strings.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "outputs" : [ - { - "name" : "Out", - "comment" : "(2-D Tensor with shape [`batch_size` x 1]) The output edit distances of EditDistance operator.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "attrs" : [ - { - "name" : "normalized", - "type" : "bool", - "comment" : "(bool, default false) Indicated whether to normalize the edit distance by the length of reference string.", - "generated" : 0 - } ] -},{ - "type" : "l1_norm", - "comment" : "\nL1 Norm Operator.\n\nComputes the L1 norm of a tensor.\n\n$$Out = \\sum{|X|}$$\n\n", - "inputs" : [ - { - "name" : "X", - "comment" : "(Tensor) The input of l1_norm op.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "outputs" : [ - { - "name" : "Out", - "comment" : "(Scalar) The output of l1_norm op.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "attrs" : [ ] -},{ - "type" : "stanh", - "comment" : "\nSTanh Activation Operator.\n\n$$out = b * \\frac{e^{a * x} - e^{-a * x}}{e^{a * x} + e^{-a * x}}$$\n\n", - "inputs" : [ - { - "name" : "X", - "comment" : "Input of STanh operator", - "duplicable" : 0, - "intermediate" : 0 - } ], - "outputs" : [ - { - "name" : "Out", - "comment" : "Output of STanh operator", - "duplicable" : 0, - "intermediate" : 0 - } ], - "attrs" : [ - { - "name" : "scale_a", - "type" : "float", - "comment" : "The scale parameter of a for the input", - "generated" : 0 - }, { - "name" : "scale_b", - "type" : "float", - "comment" : "The scale parameter of b for the input", - "generated" : 0 - } ] -},{ - "type" : "adamax", - "comment" : "\nAdamax Optimizer.\n\nWe implement the Adamax optimizer from Section 7 of the Adam\npaper: https://arxiv.org/abs/1412.6980. Adamax is a variant of the\nAdam algorithm based on the infinity norm.\n\nAdamax updates:\n\n$$\nmoment\\_out = \\beta_1 * moment + (1 - \\beta_1) * grad \\\\\ninf\\_norm\\_out = max(\\beta_2 * inf\\_norm + \\epsilon, |grad|) \\\\\nlearning\\_rate = \\frac{learning\\_rate}{1 - \\beta_{1\\_pow}} \\\\\nparam\\_out = param - learning\\_rate * \\frac{moment\\_out}{inf\\_norm\\_out}\n$$\n\nThe original paper does not have an epsilon attribute.\nHowever, it is added here for numerical stability to prevent the\ndivision by 0 error.\n\n", - "inputs" : [ - { - "name" : "Param", - "comment" : "(Tensor) Input parameter", - "duplicable" : 0, - "intermediate" : 0 - }, { - "name" : "Grad", - "comment" : "(Tensor) Input gradient", - "duplicable" : 0, - "intermediate" : 0 - }, { - "name" : "LearningRate", - "comment" : "(Tensor) Learning rate", - "duplicable" : 0, - "intermediate" : 0 - }, { - "name" : "Moment", - "comment" : "(Tensor) First moment", - "duplicable" : 0, - "intermediate" : 0 - }, { - "name" : "InfNorm", - "comment" : "(Tensor) Input exponentially weighted infinity norm", - "duplicable" : 0, - "intermediate" : 0 - }, { - "name" : "Beta1Pow", - "comment" : "(Tensor) Input beta1 power accumulator", - "duplicable" : 0, - "intermediate" : 0 - } ], - "outputs" : [ - { - "name" : "ParamOut", - "comment" : "(Tensor) Output parameter", + { + "name" : "Hyps", + "comment" : "(2-D LoDTensor, 2nd dim. equal to 1) The indices for hypothesis strings.", "duplicable" : 0, "intermediate" : 0 }, { - "name" : "MomentOut", - "comment" : "(Tensor) Output first moment", + "name" : "Refs", + "comment" : "(2-D LoDTensor, 2nd dim. equal to 1) The indices for reference strings.", "duplicable" : 0, "intermediate" : 0 - }, { - "name" : "InfNormOut", - "comment" : "(Tensor) Output exponentially weighted infinity norm", + } ], + "outputs" : [ + { + "name" : "Out", + "comment" : "(2-D Tensor with shape [`batch_size` x 1]) The output edit distances of EditDistance operator.", "duplicable" : 0, "intermediate" : 0 } ], "attrs" : [ { - "name" : "beta1", - "type" : "float", - "comment" : "(float, default 0.9) Exponential decay rate for the 1st moment estimates.", - "generated" : 0 - }, { - "name" : "beta2", - "type" : "float", - "comment" : "(float, default 0.999) exponential decay rate for the weighted infinity norm estimates.", - "generated" : 0 - }, { - "name" : "epsilon", - "type" : "float", - "comment" : "(float, default 1.0e-8) Constant for numerical stability", + "name" : "normalized", + "type" : "bool", + "comment" : "(bool, default false) Indicated whether to normalize the edit distance by the length of reference string.", "generated" : 0 } ] +},{ + "type" : "l1_norm", + "comment" : "\nL1 Norm Operator.\n\nComputes the L1 norm of a tensor.\n\n$$Out = \\sum{|X|}$$\n\n", + "inputs" : [ + { + "name" : "X", + "comment" : "(Tensor) The input of l1_norm op.", + "duplicable" : 0, + "intermediate" : 0 + } ], + "outputs" : [ + { + "name" : "Out", + "comment" : "(Scalar) The output of l1_norm op.", + "duplicable" : 0, + "intermediate" : 0 + } ], + "attrs" : [ ] },{ "type" : "swish", "comment" : "\nSwish Activation Operator.\n\n$$out = \\frac{x}{1 + e^{- \\beta x}}$$\n\n", @@ -4528,50 +4309,6 @@ "intermediate" : 0 } ], "attrs" : [ ] -},{ - "type" : "conv3d_transpose_cudnn", - "comment" : "\nConvolution3D Transpose Operator.\n\nThe convolution transpose operation calculates the output based on the input, filter\nand dilations, strides, paddings, groups parameters. The size of each dimension of the\nparameters is checked in the infer-shape.\nInput(Input) and output(Output) are in NCDHW format. Where N is batch size, C is the\nnumber of channels, D is the depth of the feature, H is the height of the feature,\nand W is the width of the feature.\nFilter(Input) is in MCDHW format. Where M is the number of input feature channels,\nC is the number of output feature channels, D is the depth of the filter,H is the\nheight of the filter, and W is the width of the filter.\nParameters(strides, paddings) are three elements. These three elements represent\ndepth, height and width, respectively.\nThe input(X) size and output(Out) size may be different.\n\nExample: \n Input:\n Input shape: $(N, C_{in}, D_{in}, H_{in}, W_{in})$\n Filter shape: $(C_{in}, C_{out}, D_f, H_f, W_f)$\n Output:\n Output shape: $(N, C_{out}, D_{out}, H_{out}, W_{out})$\n Where\n $$\n D_{out} = (D_{in} - 1) * strides[0] - 2 * paddings[0] + D_f \\\\\n H_{out} = (H_{in} - 1) * strides[1] - 2 * paddings[1] + H_f \\\\\n W_{out} = (W_{in} - 1) * strides[2] - 2 * paddings[2] + W_f\n $$\n", - "inputs" : [ - { - "name" : "Input", - "comment" : "(Tensor) The input tensor of convolution transpose operator.The format of input tensor is NCDHW. Where N is batch size, C is the number of channels, D is the depth of the feature, H is the height of the feature, and W is the width of the feature.", - "duplicable" : 0, - "intermediate" : 0 - }, { - "name" : "Filter", - "comment" : "(Tensor) The filter tensor of convolution transpose operator.The format of the filter tensor is MCDHW, where M is the number of input feature channels, C is the number of output feature channels, D is the depth of the filter, H is the height of the filter, and W is the width of the filter.We enforce groups number == 1 and padding == 0 in the convolution3d transpose scenario.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "outputs" : [ - { - "name" : "Output", - "comment" : "(Tensor) The output tensor of convolution transpose operator.The format of output tensor is also NCDHW.Where N is batch size, C is the number of channels, D is the depth of the feature, H is the height of the feature, and W is the width of the feature.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "attrs" : [ - { - "name" : "dilations", - "type" : "int array", - "comment" : "(vector default:{1, 1, 1}), the dilations(d_dilation,h_dilation, w_dilation) of convolution transpose operator.", - "generated" : 0 - }, { - "name" : "strides", - "type" : "int array", - "comment" : "(vector default:{1, 1, 1}), the strides{d_stride, h_stride, w_stride} of convolution transpose operator.", - "generated" : 0 - }, { - "name" : "paddings", - "type" : "int array", - "comment" : "(vector default:{0, 0, 0}), paddings(d_pad, h_pad, w_pad) of convolution transpose operator.", - "generated" : 0 - }, { - "name" : "workspace_size_MB", - "type" : "int", - "comment" : "workspace size for cudnn, in MB, workspace is a section of GPU memory which will be allocated/freed each time the operator runs, larger workspace size can increase performance but also requires better hardward. This size should be carefully setted.", - "generated" : 0 - } ] },{ "type" : "conv_shift", "comment" : "\nConvShift Operator.\n\nA layer for circular convolution of two vectors,\nas used in the Neural Turing Machine: https://arxiv.org/abs/1410.5401\n\nThe equation is:\n\n$$Out[i] = \\sum_{j=-(N-1)/2}^{(N-1)/2} X_{i+j} * Y_{j}$$\n\nwhere X's index is computed modulo M, and Y's index is computed modulo N.\n\nBoth inputs X and Y can carry LoD (Level of Details) information.\nHowever, the output only shares the LoD information with input X.\n\n", @@ -4595,55 +4332,6 @@ "intermediate" : 0 } ], "attrs" : [ ] -},{ - "type" : "conv2d_cudnn", - "comment" : "\nConvolution Operator.\n\nThe convolution operation calculates the output based on the input, filter\nand strides, paddings, dilations, groups parameters. The size of each dimension of the\nparameters is checked in the infer-shape.\nInput(Input) and Output(Output) are in NCHW format. Where N is batch\nsize, C is the number of channels, H is the height of the feature, and W is\nthe width of the feature.\nFilters(Input) is MCHW format. Where M is the number of output image channels, C is\nthe number of input image channels, H is the height of the filter, and W\nis the width of the filter.\nParameters(strides, paddings, dilations) are two elements. These two elements represent\nheight and width, respectively.\nThe input(X) size and output(Out) size may be different.\n\nExample:\n Input:\n Input shape: $(N, C_{in}, H_{in}, W_{in})$\n Filter shape: $(C_{out}, C_{in}, H_f, W_f)$\n Output:\n Output shape: $(N, C_{out}, H_{out}, W_{out})$\n Where\n$$\n H_{out}= \\frac{(H_{in} + 2 * paddings[0] - (dilations[0] * (H_f - 1) + 1))}{strides[0]}+ 1 \\\\\n W_{out}= \\frac{(W_{in} + 2 * paddings[1] - (dilations[1] * (W_f - 1) + 1))}{strides[1]}+ 1\n$$\n", - "inputs" : [ - { - "name" : "Input", - "comment" : "(Tensor) The input tensor of convolution operator. The format of input tensor is NCHW, where N is batch size, C is the number of channels, H is the height of the feature, and W is the width of the feature.", - "duplicable" : 0, - "intermediate" : 0 - }, { - "name" : "Filter", - "comment" : "(Tensor) The filter tensor of convolution operator. The format of the filter tensor is MCHW, where M is the number of output image channels, C is the number of input image channels, H is the height of the filter, and W is the width of the filter. If the groups attribute is greater than 1, C equals the number of input image channels divided by the groups.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "outputs" : [ - { - "name" : "Output", - "comment" : "(Tensor) The output tensor of convolution operator. The format of output tensor is also NCHW.", - "duplicable" : 0, - "intermediate" : 0 - } ], - "attrs" : [ - { - "name" : "strides", - "type" : "int array", - "comment" : "(vector default:{1, 1}), the strides(h_stride, w_stride) of convolution operator.", - "generated" : 0 - }, { - "name" : "paddings", - "type" : "int array", - "comment" : "(vector default:{0, 0}), the paddings(h_pad, w_pad) of convolution operator.", - "generated" : 0 - }, { - "name" : "groups", - "type" : "int", - "comment" : "(int default:1), the groups number of the convolution operator. According to grouped convolution in Alex Krizhevsky's Deep CNN paper: when group=2, the first half of the filters is only connected to the first half of the input channels, while the second half of the filters is only connected to the second half of the input channels.", - "generated" : 0 - }, { - "name" : "dilations", - "type" : "int array", - "comment" : "(vector default:{1, 1}), the dilations(h_dilation, w_dilation) of convolution operator.", - "generated" : 0 - }, { - "name" : "workspace_size_MB", - "type" : "int", - "comment" : "workspace size for cudnn, in MB, workspace is a section of GPU memory which will be allocated/freed each time the operator runs, larger workspace size can increase performance but also requires better hardware. This size should be chosen carefully.", - "generated" : 0 - } ] },{ "type" : "conditional_block", "comment" : "Conditional block operator\n\nRun the sub-block if X is not empty. Params is the other inputs and Out is the\noutputs of the sub-block.\n", @@ -5184,6 +4872,70 @@ "comment" : "The number of thresholds to use when discretizing the roc curve.", "generated" : 0 } ] +},{ + "type" : "get_places", + "comment" : "\nReturns a list of places based on flags. The list will be used for parallel\nexecution.\n", + "inputs" : [ ], + "outputs" : [ + { + "name" : "Out", + "comment" : "vector of Place", + "duplicable" : 0, + "intermediate" : 0 + } ], + "attrs" : [ + { + "name" : "device_count", + "type" : "int", + "comment" : "device count", + "generated" : 0 + }, { + "name" : "device_type", + "type" : "string", + "comment" : "device type", + "generated" : 0 + } ] +},{ + "type" : "read_from_array", + "comment" : "\nReadFromArray Operator.\n\nRead a LoDTensor from a LoDTensor Array.\n\nAssume $T$ is LoDTensor, $i$ is the subscript of the array, and $A$ is the array. The\nequation is\n\n$$T = A[i]$$\n\n", + "inputs" : [ + { + "name" : "X", + "comment" : "(TensorArray) the array will be read from.", + "duplicable" : 0, + "intermediate" : 0 + }, { + "name" : "I", + "comment" : "(Tensor) the subscript index in tensor array. The number of element should be 1", + "duplicable" : 0, + "intermediate" : 0 + } ], + "outputs" : [ + { + "name" : "Out", + "comment" : "(LoDTensor) the tensor will be read from.", + "duplicable" : 0, + "intermediate" : 0 + } ], + "attrs" : [ ] +},{ + "type" : "softplus", + "comment" : "\nSoftplus Activation Operator.\n\n$out = \\ln(1 + e^{x})$\n\n", + "inputs" : [ + { + "name" : "X", + "comment" : "Input of Softplus operator", + "duplicable" : 0, + "intermediate" : 0 + } ], + "outputs" : [ + { + "name" : "Out", + "comment" : "Output of Softplus operator", + "duplicable" : 0, + "intermediate" : 0 + } ], + "attrs" : [ ] },{ "type" : "assign_value", "comment" : "\nAssignValue operator\n\n$$Out = values$$\n", @@ -5425,6 +5177,24 @@ "comment" : "(float, default 1.0e-6) Constant for numerical stability", "generated" : 0 } ] +},{ + "type" : "log", + "comment" : "\nLog Activation Operator.\n\n$out = \\ln(x)$\n\nNatural logarithm of x.\n\n", + "inputs" : [ + { + "name" : "X", + "comment" : "Input of Log operator", + "duplicable" : 0, + "intermediate" : 0 + } ], + "outputs" : [ + { + "name" : "Out", + "comment" : "Output of Log operator", + "duplicable" : 0, + "intermediate" : 0 + } ], + "attrs" : [ ] },{ "type" : "nce", "comment" : "\nCompute and return the noise-contrastive estimation training loss.\nSee [Noise-contrastive estimation: A new estimation principle for unnormalized statistical models](http://www.jmlr.org/proceedings/papers/v9/gutmann10a/gutmann10a.pdf).\nBy default this operator uses a uniform distribution for sampling.\n", @@ -5727,4 +5497,40 @@ "comment" : "(float, default -0.5f) Learning Rate Power.", "generated" : 0 } ] +},{ + "type" : "round", + "comment" : "\nRound Activation Operator.\n\n$out = [x]$\n\n", + "inputs" : [ + { + "name" : "X", + "comment" : "Input of Round operator", + "duplicable" : 0, + "intermediate" : 0 + } ], + "outputs" : [ + { + "name" : "Out", + "comment" : "Output of Round operator", + "duplicable" : 0, + "intermediate" : 0 + } ], + "attrs" : [ ] +},{ + "type" : "softsign", + "comment" : "\nSoftsign Activation Operator.\n\n$$out = \\frac{x}{1 + |x|}$$\n\n", + "inputs" : [ + { + "name" : "X", + "comment" : "Input of Softsign operator", + "duplicable" : 0, + "intermediate" : 0 + } ], + "outputs" : [ + { + "name" : "Out", + "comment" : "Output of Softsign operator", + "duplicable" : 0, + "intermediate" : 0 + } ], + "attrs" : [ ] }] diff --git a/develop/doc_cn/_sources/howto/dev/new_op_cn.md.txt b/develop/doc_cn/_sources/howto/dev/new_op_cn.md.txt index 3109d72001..9299658567 100644 --- a/develop/doc_cn/_sources/howto/dev/new_op_cn.md.txt +++ b/develop/doc_cn/_sources/howto/dev/new_op_cn.md.txt @@ -24,7 +24,7 @@ - `framework::OperatorWithKernel`：继承自OperatorBase，Op有计算函数，称作有Kernel。 - `class OpProtoAndCheckerMaker`：描述该Op的输入、输出、属性、注释,主要用于Python API接口生成 -依据是否包含kernel，可以将Op分为两种：包含Kernel的Op和不包含kernel的Op，前者Op的定义继承自`OperatorBase`，后者继承自`OperatorWithKernel`。本教程主要介绍带Kernel的Op如何写，简单总结Op需要包含的内容如下： +依据是否包含kernel，可以将Op分为两种：包含Kernel的Op和不包含kernel的Op，前者Op的定义继承自`OperatorWithKernel`，后者继承自`OperatorBase`。本教程主要介绍带Kernel的Op如何写，简单总结Op需要包含的内容如下： 内容 | 定义位置 diff --git a/develop/doc_cn/howto/dev/new_op_cn.html b/develop/doc_cn/howto/dev/new_op_cn.html index 6742d514ff..88f526efb6 100644 --- a/develop/doc_cn/howto/dev/new_op_cn.html +++ b/develop/doc_cn/howto/dev/new_op_cn.html @@ -258,7 +258,7 @@
  • framework::OperatorWithKernel：继承自OperatorBase，Op有计算函数，称作有Kernel。
  • class OpProtoAndCheckerMaker：描述该Op的输入、输出、属性、注释,主要用于Python API接口生成
    • -

    依据是否包含kernel，可以将Op分为两种：包含Kernel的Op和不包含kernel的Op，前者Op的定义继承自OperatorBase，后者继承自OperatorWithKernel。本教程主要介绍带Kernel的Op如何写，简单总结Op需要包含的内容如下：

    +

    依据是否包含kernel，可以将Op分为两种：包含Kernel的Op和不包含kernel的Op，前者Op的定义继承自OperatorWithKernel，后者继承自OperatorBase。本教程主要介绍带Kernel的Op如何写，简单总结Op需要包含的内容如下：

    内容 | 定义位置 ————– | :———————- OpProtoMake定义 | .cc文件，Backward Op不需要定义OpProtoMake -- GitLab