# bidirectionalLSTM(initialHiddenStates:initialCellStates:inputHiddenWeight:hiddenHiddenWeight:bias:inputHiddenWeightBack:hiddenHiddenWeightBack:biasBack:activation:recurrentActivation:cellActivation:outputSequence:) Adds a bidirectional LSTM operation to the current graph. ``` func bidirectionalLSTM(initialHiddenStates: BNNSGraph.Builder.Tensor, initialCellStates: BNNSGraph.Builder.Tensor, inputHiddenWeight: BNNSGraph.Builder.Tensor, hiddenHiddenWeight: BNNSGraph.Builder.Tensor, bias: BNNSGraph.Builder.Tensor, inputHiddenWeightBack: BNNSGraph.Builder.Tensor, hiddenHiddenWeightBack: BNNSGraph.Builder.Tensor, biasBack: BNNSGraph.Builder.Tensor, activation: BNNSGraph.Builder.Activation, recurrentActivation: BNNSGraph.Builder.Activation, cellActivation: BNNSGraph.Builder.Activation, outputSequence: Bool) -> (output: BNNSGraph.Builder.Tensor, hiddenStates: BNNSGraph.Builder.Tensor, memoryStates: BNNSGraph.Builder.Tensor) ``` ## Parameters `initialHiddenStates` The initial hidden states with the shape `(N, 2*Hout)`. `initialCellStates` The initial hidden states with the shape `(N, 2*Hout)`. `inputHiddenWeight` The input-hidden weight with the shape `(4*Hout, Hin)`. `bias` The bias (the sum of input-hidden and hidden-hidden biases) with the shape `(4*Hout,)`. `inputHiddenWeightBack` The backward input-hidden weight with the shape `(4*Hout, Hin)`. `hiddenHiddenWeightBack` The backward hidden-hidden weight with the shape `(4*Hout, Hout)`. `biasBack` The backward bias (the sum of input-hidden and hidden-hidden biases) with the shape`(4*Hout,)`. `activation` An enumeration that controls the output activation function. `recurrentActivation` An enumeration that controls the recurrent activation function. `cellActivation` An enumeration that controls the cell activation function. `outputSequence` When `true`, `output` is of shape `(L, N, 2*Hout)` and contains hidden states from every step, `h[:, ...]`. When `false`, ` output` is of shape `(1, N, 2*Hout)` and contains hidden states from the last step, `h[-1, ...]`. ## Discussion The input tensor `x` is of shape `(L, N, Hin)` - Parameter hiddenHiddenWeight The hidden-hidden weight with the shape `(4*Hout, Hout)`. `hiddenStates` is of shape `(N, 2*Hout)` and contains hidden states from the last step, `h[-1, ...]` `memoryStates` is of shape `(N, 2*Hout)` and contains memory states from the last step, `c[-1, ...]` > seealso: ``doc://com.apple.accelerate/documentation/Accelerate/BNNSGraph/Builder/Tensor/lstm(initialHiddenStates:initialCellStates:inputHiddenWeight:hiddenHiddenWeight:bias:direction:activation:recurrentActivation:cellActivation:outputSequence:)`` --- Copyright © 2026 Apple Inc. All rights reserved. | [Terms of Use](https://www.apple.com/legal/internet-services/terms/site.html) | [Privacy Policy](https://www.apple.com/privacy/privacy-policy)