Decoder

class opennmt.decoders.Decoder(*args, **kwargs)[source]

Base class for decoders.

Inherits from: keras.src.engine.base_layer.Layer

Extended by:

__init__(num_sources=1, vocab_size=None, output_layer=None, output_layer_bias=True, **kwargs)[source]

Initializes the decoder parameters.

If you don’t set one of vocab_size or output_layer here, you should later call the method opennmt.decoders.Decoder.initialize() to initialize this decoder instance.

Parameters

num_sources – The number of source contexts expected by this decoder.
vocab_size – The output vocabulary size (optional if output_layer is set).
output_layer – The output projection layer (optional).
output_layer_bias – Add bias after the output projection layer.
**kwargs – Additional layer arguments.

Raises

ValueError – if the number of source contexts num_sources is not supported by this decoder.

property minimum_sources: The minimum number of source contexts supported by this decoder.

property maximum_sources: The maximum number of source contexts supported by this decoder.

property support_alignment_history: Returns True if this decoder can return the attention as alignment history.

property initialized: Returns True if this decoder is initialized.

initialize(vocab_size=None, output_layer=None)[source]

Initializes the decoder configuration.

Parameters

vocab_size – The target vocabulary size.
output_layer – The output layer to use.

Raises

ValueError – if both vocab_size and output_layer are not set.

reuse_embeddings(embeddings)[source]

Reuses embeddings in the decoder output layer.

Parameters: embeddings – The embeddings matrix to reuse.
Raises: RuntimeError – if the decoder was not initialized.

initial_state(memory=None, memory_sequence_length=None, initial_state=None, batch_size=None, dtype=None)[source]

Returns the initial decoder state.

Parameters

memory – Memory values to query.
memory_sequence_length – Memory values length.
initial_state – An initial state to start from, e.g. the last encoder state.
batch_size – The batch size to use.
dtype – The dtype of the state.

Returns

A nested structure of tensors representing the decoder state.

Raises

RuntimeError – if the decoder was not initialized.
ValueError – if one of batch_size or dtype is not set and neither initial_state nor memory are not passed.
ValueError – if the number of source contexts (memory) does not match the number defined at the decoder initialization.

call(inputs, length_or_step=None, state=None, input_fn=None, sampling_probability=None, training=None)[source]

Runs the decoder layer on either a complete sequence (e.g. for training or scoring), or a single timestep (e.g. for iterative decoding).

Parameters

inputs – The inputs to decode, can be a 3D (training) or 2D (iterative decoding) tensor.
length_or_step – For 3D inputs, the length of each sequence. For 2D inputs, the current decoding timestep.
state – The decoder state.
input_fn – A callable taking sampled ids and returning the decoding inputs.
sampling_probability – When inputs is the full sequence, the probability to read from the last sample instead of the true target.
training – Run in training mode.

Returns

A tuple with the logits, the decoder state, and an attention vector.

Raises

RuntimeError – if the decoder was not initialized.
ValueError – if the inputs rank is different than 2 or 3.
ValueError – if length_or_step is invalid.

forward(inputs, sequence_length=None, initial_state=None, memory=None, memory_sequence_length=None, input_fn=None, sampling_probability=None, training=None)[source]

Runs the decoder on full sequences.

Parameters

inputs – The 3D decoder input.
sequence_length – The length of each input sequence.
initial_state – The initial decoder state.
memory – Memory values to query.
memory_sequence_length – Memory values length.
input_fn – A callable taking sampled ids and returning the decoding inputs.
sampling_probability – The probability to read from the last sample instead of the true target.
training – Run in training mode.

Returns

A tuple with the logits, the decoder state, and the attention vector.

abstract step(inputs, timestep, state=None, memory=None, memory_sequence_length=None, training=None)[source]

Runs one decoding step.

Parameters

inputs – The 2D decoder input.
timestep – The current decoding step.
state – The decoder state.
memory – Memory values to query.
memory_sequence_length – Memory values length.
training – Run in training mode.

Returns

A tuple with the decoder outputs, the decoder state, and the attention vector.

dynamic_decode(embeddings, start_ids, end_id=2, initial_state=None, decoding_strategy=None, sampler=None, maximum_iterations=None, minimum_iterations=0, tflite_output_size=None)[source]

Decodes dynamically from start_ids.

Parameters

embeddings – Target embeddings or opennmt.inputters.WordEmbedder to apply on decoded ids.
start_ids – Initial input IDs of shape \([B]\).
end_id – ID of the end of sequence token.
initial_state – Initial decoder state.
decoding_strategy – A opennmt.utils.DecodingStrategy instance that define the decoding logic. Defaults to a greedy search.
sampler – A opennmt.utils.Sampler instance that samples predictions from the model output. Defaults to an argmax sampling.
maximum_iterations – The maximum number of iterations to decode for.
minimum_iterations – The minimum number of iterations to decode for.
tflite_output_size – If not None will run TFLite safe, is the size of 1D output tensor.

Returns

A opennmt.utils.DecodingResult instance.