RNN - 1 vs 22

Next section compares an older to a newer version of the same operator after both definition are converted into markdown text. Green means an addition to the newer version, red means a deletion. Anything else is unchanged.

Files changed (1) hide show

1. RNN1 → RNN22 +31 -51

RNN1 → RNN22 RENAMED Viewed

	@@ -1 +1 @@
1 1	Computes an one-layer simple RNN. This operator is usually supported
2 2	via some custom implementation such as CuDNN.
3 3	Notations:
4	- X - input tensor
4	+ * X - input tensor
5	-
6	- i - input gate
5	+ * i - input gate
7	-
8	- t - time step (t-1 means previous time step)
6	+ * t - time step (t-1 means previous time step)
9	-
10	- Wi - W parameter weight matrix for input gate
7	+ * Wi - W parameter weight matrix for input gate
11	-
12	- Ri - R recurrence weight matrix for input gate
8	+ * Ri - R recurrence weight matrix for input gate
13	-
14	- Wbi - W parameter bias vector for input gate
9	+ * Wbi - W parameter bias vector for input gate
15	-
16	- Rbi - R parameter bias vector for input gate
10	+ * Rbi - R parameter bias vector for input gate
17	-
18	- WBi - W parameter weight matrix for backward input gate
11	+ * WBi - W parameter weight matrix for backward input gate
19	-
20	- RBi - R recurrence weight matrix for backward input gate
12	+ * RBi - R recurrence weight matrix for backward input gate
21	-
22	- WBbi - WR bias vectors for backward input gate
13	+ * WBbi - WR bias vectors for backward input gate
23	-
24	- RBbi - RR bias vectors for backward input gate
14	+ * RBbi - RR bias vectors for backward input gate
25	-
26	- H - Hidden state
15	+ * H - Hidden state
27	-
28	- num_directions - 2 if direction == bidirectional else 1
16	+ * num_directions - 2 if direction == bidirectional else 1
29 17	Activation functions:
30	- Relu(x) - max(0, x)
18	+ * Relu(x) - max(0, x)
19	+ * Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x})
20	+ * Sigmoid(x) - 1/(1 + e^{-x})
31	- Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x})
21	+ NOTE: Below are optional
32	- Sigmoid(x) - 1/(1 + e^{-x})
33	-
34	- (NOTE: Below are optional)
35	-
36	- Affine(x) - alpha*x + beta
22	+ * Affine(x) - alpha*x + beta
37	-
38	- LeakyRelu(x) - x if x >= 0 else alpha * x
23	+ * LeakyRelu(x) - x if x >= 0 else alpha * x
39	-
40	- ThresholdedRelu(x) - x if x >= alpha else 0
24	+ * ThresholdedRelu(x) - x if x >= alpha else 0
41	-
42	- ScaledTanh(x) - alpha*Tanh(beta*x)
25	+ * ScaledTanh(x) - alpha*Tanh(beta*x)
43	-
44	- HardSigmoid(x) - min(max(alpha*x + beta, 0), 1)
26	+ * HardSigmoid(x) - min(max(alpha*x + beta, 0), 1)
45	-
46	- Elu(x) - x if x >= 0 else alpha*(e^x - 1)
27	+ * Elu(x) - x if x >= 0 else alpha*(e^x - 1)
47	-
48	- Softsign(x) - x/(1 + |x|)
28	+ * Softsign(x) - x/(1 + |x|)
49	-
50	- Softplus(x) - log(1 + e^x)
29	+ * Softplus(x) - log(1 + e^x)
51 30	Equations (Default: f=Tanh):
52	- - Ht = f(Xt*(Wi^T) + Ht-1*Ri + Wbi + Rbi)
31	+ * Ht = f(Xt*(Wi^T) + Ht-1*(Ri^T) + Wbi + Rbi)
32	+ This operator has **optional** inputs/outputs. See [ONNX IR](https://github.com/onnx/onnx/blob/main/docs/IR.md) for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted.
53 33	### Attributes
54 34	* **activation_alpha - FLOATS** :
55 35	Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01.
56 36	* **activation_beta - FLOATS** :
57 37	Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.
58 38	* **activations - STRINGS** (default is ['Tanh', 'Tanh']):
59 39	One (or two if bidirectional) activation function for input gate. The activation function must be one of the activation functions specified above. Optional: Default Tanh if not specified.
60 40	* **clip - FLOAT** :
61 41	Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
62 42	* **direction - STRING** (default is 'forward'):
63 43	Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional.
64 44	* **hidden_size - INT** :
65 45	Number of neurons in the hidden layer
66	- * **output_sequence - INT** (default is '0'):
46	+ * **layout - INT** (default is '0'):
67	- The sequence output for the hidden is optional if 0. Default 0.
47	+ The shape format of inputs X, initial_h and outputs Y, Y_h. If 0, the following shapes are expected: X.shape = [seq_length, batch_size, input_size], Y.shape = [seq_length, num_directions, batch_size, hidden_size], initial_h.shape = Y_h.shape = [num_directions, batch_size, hidden_size]. If 1, the following shapes are expected: X.shape = [batch_size, seq_length, input_size], Y.shape = [batch_size, seq_length, num_directions, hidden_size], initial_h.shape = Y_h.shape = [batch_size, num_directions, hidden_size].
68 48	### Inputs
69 49	Between 3 and 6 inputs.
70 50	- **X** (heterogeneous) - **T**:
71 51	The input sequences packed (and potentially padded) into one 3-D tensor with the shape of [seq_length, batch_size, input_size].
72 52	- **W** (heterogeneous) - **T**:
73 53	The weight tensor for input gate. Concatenation of Wi and WBi (if bidirectional). The tensor has shape [num_directions, hidden_size, input_size].
74 54	- **R** (heterogeneous) - **T**:
75 55	The recurrence weight tensor. Concatenation of Ri and RBi (if bidirectional). The tensor has shape [num_directions, hidden_size, hidden_size].
76 56	- **B** (optional, heterogeneous) - **T**:
77 57	The bias tensor for input gate. Concatenation of [Wbi, Rbi] and [WBbi, RBbi] (if bidirectional). The tensor has shape [num_directions, 2*hidden_size]. Optional: If not specified - assumed to be 0.
78 58	- **sequence_lens** (optional, heterogeneous) - **T1**:
79 59	Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length seq_length. It has shape [batch_size].
80 60	- **initial_h** (optional, heterogeneous) - **T**:
81 61	Optional initial value of the hidden. If not specified - assumed to be 0. It has shape [num_directions, batch_size, hidden_size].
82 62	### Outputs
83 63	Between 0 and 2 outputs.
84 64	- **Y** (optional, heterogeneous) - **T**:
85	- A tensor that concats all the intermediate output values of the hidden. It has shape [seq_length, num_directions, batch_size, hidden_size]. It is optional if output_sequence is 0.
65	+ A tensor that concats all the intermediate output values of the hidden. It has shape [seq_length, num_directions, batch_size, hidden_size].
86 66	- **Y_h** (optional, heterogeneous) - **T**:
87 67	The last output value of the hidden. It has shape [num_directions, batch_size, hidden_size].
88 68	### Type Constraints
89	- * **T** in ( tensor(double), tensor(float), tensor(float16) ):
69	+ * **T** in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ):
90 70	Constrain input and output types to float tensors.
91 71	* **T1** in ( tensor(int32) ):
92 72	Constrain seq_lens to integer tensor.