Struct Lstm 

pub struct Lstm {
    pub input_gate: GateController,
    pub forget_gate: Option<GateController>,
    pub output_gate: GateController,
    pub cell_gate: GateController,
    pub d_hidden: usize,
    pub batch_first: bool,
    pub reverse: bool,
    pub clip: Option<f64>,
    pub input_forget: bool,
    pub gate_activation: Activation,
    pub cell_activation: Activation,
    pub hidden_activation: Activation,
}

The Lstm module. This implementation is for a unidirectional, stateless, Lstm.

Introduced in the paper: Long Short-Term Memory.

Should be created with LstmConfig.

Checkpoint compatibility

Initialize with the same input_forget setting used during training before loading weights. Older coupled LSTM checkpoints contain unused forget-gate tensors. Load these with ModuleRecord::allow_unused(true), then save the module again to omit those tensors. Uncoupled checkpoints retain the same parameter paths.

Fields

input_gate: GateController

The input gate regulates which information to update and store in the cell state at each time step.

forget_gate: Option<GateController>

The forget gate is used to control which information to discard or keep in the memory cell at each time step. Absent when input_forget is true (forget = 1 - input). Access an uncoupled gate with as_ref() or as_mut(), and replace it with Some(gate).

output_gate: GateController

The output gate determines which information from the cell state to output at each time step.

cell_gate: GateController

The cell gate is used to compute the cell state that stores and carries information through time.

d_hidden: usize

The hidden state of the LSTM.

batch_first: bool

If true, input is [batch_size, seq_length, input_size]. If false, input is [seq_length, batch_size, input_size].

reverse: bool

If true, process the sequence in reverse order.

clip: Option<f64>

Optional cell state clip threshold.

input_forget: bool

If true, couples input and forget gates: f_t = 1 - i_t. Set through LstmConfig::input_forget at initialization so the forget gate is omitted.

gate_activation: Activation

Activation function for gates (input, forget, output).

cell_activation: Activation

Activation function for cell gate (candidate cell state).

hidden_activation: Activation

Activation function for hidden output.

Implementations

impl Lstm

pub fn forward( &self, batched_input: Tensor<3>, state: Option<LstmState<2>>, ) -> (Tensor<3>, LstmState<2>)

Applies the forward pass on the input tensor. This LSTM implementation returns the state for each element in a sequence (i.e., across seq_length) and a final state.

Parameters:

• batched_input: The input tensor of shape:
  • [batch_size, sequence_length, input_size] if batch_first is true (default)
  • [sequence_length, batch_size, input_size] if batch_first is false
• state: An optional LstmState representing the initial cell state and hidden state. Each state tensor has shape [batch_size, hidden_size]. If no initial state is provided, these tensors are initialized to zeros.

Returns:

• output: A tensor represents the output features of LSTM. Shape:
  • [batch_size, sequence_length, hidden_size] if batch_first is true
  • [sequence_length, batch_size, hidden_size] if batch_first is false
• state: A LstmState represents the final states. Both state.cell and state.hidden have the shape [batch_size, hidden_size].

pub fn forward_iter<I>( &self, input_timestep_iter: I, state: Option<LstmState<2>>, batch_size: usize, seq_length: usize, device: &Device, ) -> (Tensor<3>, LstmState<2>)

where I: Iterator<Item = (usize, Tensor<3>)>,

Applies the forward iteration over input timesteps for the LSTM. This method is called internally by the forward function to process each timestep of the sequence.

Parameters:

• input_timestep_iter: An iterator where each item is a pair consisting of the timestep index (usize) and the corresponding input tensor of shape [batch_size, 1, d_input].
• state: An optional LstmState representing the initial cell and hidden states. If None, the states are initialized to zeros with shapes [batch_size, d_hidden].
• batch_size: The number of sequences in the batch.
• seq_length: The length of the input sequence (number of timesteps).
• device: The device where computations will run.

Returns:

• A pair where:
  • The first element is a tensor of shape [batch_size, seq_length, d_hidden], containing the output features for all timesteps.
  • The second element is the final LstmState, which includes the final cell and hidden states, both with the shape [batch_size, d_hidden].

Trait Implementations

impl Clone for Lstm

fn clone(&self) -> Lstm

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for Lstm

fn fmt(&self, f: &mut Formatter<'_>) -> Result<(), Error>

Formats the value using the given formatter.

impl Display for Lstm

fn fmt(&self, f: &mut Formatter<'_>) -> Result<(), Error>

Formats the value using the given formatter.

impl Module for Lstm

fn num_params(&self) -> usize

Get the number of parameters the module has, including all of its sub-modules, without initializing the ones not initialized yet.

fn visit<Visitor>(&self, visitor: &mut Visitor)

where Visitor: ModuleVisitor,

Visit each parameter and module-owned control value with a visitor.

fn map<Mapper>(self, mapper: &mut Mapper) -> Lstm

where Mapper: ModuleMapper,

Map each parameter and module-owned control value with a mapper.

fn collect_devices(&self, devices: Vec<Device>) -> Vec<Device>

Return all the devices found in the underneath module tree added to the given vector without duplicates.

fn to_device(self, device: &Device) -> Lstm

Move the module and all of its sub-modules to the given device.

fn fork(self, device: &Device) -> Lstm

Fork the module and all of its sub-modules to the given device.

fn valid(&self) -> Lstm

Returns a validation snapshot without autodiff or active training flags.

fn train(self) -> Lstm

Enables autodiff and restores configured training state throughout the module.

fn materialize(self) -> Lstm

Fold parameter reparameterizations into their effective values and remove them.

fn devices(&self) -> Vec<Device>

Return all the devices found in the underneath module tree without duplicates.

fn set_require_grad(self, require_grad: bool) -> Self

Set whether every floating-point tensor parameter in the module tree requires gradients.

fn set_require_grad_group(self, group: ParamGroup, require_grad: bool) -> Self

Set whether matched floating-point tensor parameters require gradients.

fn no_grad(self) -> Self

Disable gradients for every floating-point tensor parameter in the module tree.

fn freeze(self) -> Self

Freeze the whole module tree.

fn unfreeze(self) -> Self

Unfreeze the whole module tree.

fn freeze_group(self, group: ParamGroup) -> Self

Freeze every module-owned value in the given group, leaving the rest of the module untouched.

fn unfreeze_group(self, group: ParamGroup) -> Self

Unfreeze every module-owned value in the given group, leaving the rest of the module untouched.

fn quantize_weights(self, quantizer: &mut Quantizer) -> Self

Quantize the weights of the module.

fn quantize_weights_group( self, quantizer: &mut Quantizer, group: ParamGroup, ) -> Self

Quantize the weights of the given parameter group.

fn apply_reparameterization<R>(self, reparameterizer: R) -> Self

where Self: Sized, R: Reparameterizer,

Attach reparameterizations using the given Reparameterizer.

fn apply_lora(self, lora: Lora) -> Self

where Self: Sized,

Attach LoRA adapters to the module’s 2-D weights, freezing all base tensor parameters.

fn apply_qlora(self, qlora: QLora) -> Self

where Self: Sized,

Apply QLoRA to the module: quantize the (frozen) base tensor parameters and attach trainable LoRA adapters to 2-D weights.

fn into_record(self) -> ModuleRecord

where Self: Sized,

Collect this module’s tensor parameters into a ModuleRecord.

fn into_record_group(self, group: ParamGroup) -> ModuleRecord

where Self: Sized,

Collect the tensor parameters group names into a ModuleRecord.

fn try_load_record(self, record: ModuleRecord) -> Result<Self, RecordError>

where Self: Sized,

Apply a ModuleRecord to this module, returning the loaded module.

fn load_record(self, record: ModuleRecord) -> Self

where Self: Sized,

Apply a ModuleRecord to this module, consuming and returning it.

fn save_file<P>(self, path: P) -> Result<(), RecordError>

where P: AsRef<Path>, Self: Sized,

Save this module’s parameters to a burnpack file on disk.

fn load_file<P>(self, path: P) -> Self

where P: AsRef<Path>, Self: Sized,

Load this module’s parameters from a burnpack file on disk, returning the loaded module.

fn try_load_file<P>(self, path: P) -> Result<Self, RecordError>

where P: AsRef<Path>, Self: Sized,

Fallible variant of load_file.

impl ModuleDisplay for Lstm

fn custom_settings(&self) -> Option<DisplaySettings>

Custom display settings for the module.

fn custom_content(&self, content: Content) -> Option<Content>

Custom attributes for the module.

fn format(&self, passed_settings: DisplaySettings) -> String

Formats the module with provided display settings.

impl ModuleDisplayDefault for Lstm

fn content(&self, content: Content) -> Option<Content>

Attributes of the module used for display purposes.

fn num_params(&self) -> usize

Gets the number of the parameters of the module.