LSTM

Description

Computes a one-layer LSTM.

Input

Three to eight inputs:

X: tensor of type float16.

W: tensor of type float16.

R: tensor of type float16.

B: tensor of type float16.

sequence_lens: tensor of type int32.

initial_h: tensor of type float16.

initial_c: tensor of type float16.

p: tensor of type float16.

Output

Three outputs:

Y: tensor of type float16.

Y_h: tensor of type float16.

Y_c: tensor of type float16.

Attribute

direction: string. Defaults to forward.

hidden_size: int.

activation_alpha: (optional) scaling value used by some activation functions. The data type is float.

activation_beta: (optional) scaling value used by some activation functions. The data type can be float.

clip: Crops the elements of a tensor to the range of [-clip, +clip] and applies to the activated input. By default, no tailoring is performed. The data type can be float.

input_forget: The default value is 0. If the value is 1, the input and forget gate are coupled. Data type: int32.

ONNX Opset Support

Opset v8/v9/v10/v11/v12/v13/v14/v15/v16