Transformers documentation
인코더-디코더 모델
인코더-디코더 모델
개요
EncoderDecoderModel은 사전 학습된 자동 인코딩(autoencoding) 모델을 인코더로, 사전 학습된 자가 회귀(autoregressive) 모델을 디코더로 활용하여 시퀀스-투-시퀀스(sequence-to-sequence) 모델을 초기화하는 데 이용됩니다.
사전 학습된 체크포인트를 활용해 시퀀스-투-시퀀스 모델을 초기화하는 것이 시퀀스 생성(sequence generation) 작업에 효과적이라는 점이 Sascha Rothe, Shashi Narayan, Aliaksei Severyn의 논문 Leveraging Pre-trained Checkpoints for Sequence Generation Tasks에서 입증되었습니다.
EncoderDecoderModel이 학습/미세 조정된 후에는 다른 모델과 마찬가지로 저장/불러오기가 가능합니다. 자세한 사용법은 예제를 참고하세요.
이 아키텍처의 한 가지 응용 사례는 두 개의 사전 학습된 BertModel을 각각 인코더와 디코더로 활용하여 요약 모델(summarization model)을 구축하는 것입니다. 이는 Yang Liu와 Mirella Lapata의 논문 Text Summarization with Pretrained Encoders에서 제시된 바 있습니다.
모델 설정에서 EncoderDecoderModel 을 무작위 초기화하기
EncoderDecoderModel은 인코더와 디코더 설정(config)을 기반으로 무작위 초기화를 할 수 있습니다. 아래 예시는 BertModel 설정을 인코더로, 기본 BertForCausalLM 설정을 디코더로 사용하는 방법을 보여줍니다.
>>> from transformers import BertConfig, EncoderDecoderConfig, EncoderDecoderModel
>>> config_encoder = BertConfig()
>>> config_decoder = BertConfig()
>>> config = EncoderDecoderConfig.from_encoder_decoder_configs(config_encoder, config_decoder)
>>> model = EncoderDecoderModel(config=config)사전 학습된 인코더와 디코더로 EncoderDecoderModel 초기화하기
EncoderDecoderModel은 사전 학습된 인코더 체크포인트와 사전 학습된 디코더 체크포인트를 사용해 초기화할 수 있습니다. BERT와 같은 모든 사전 학습된 자동 인코딩(auto-encoding) 모델은 인코더로 활용할 수 있으며, GPT2와 같은 자가 회귀(autoregressive) 모델이나 BART의 디코더와 같이 사전 학습된 시퀀스-투-시퀀스 디코더 모델을 디코더로 사용할 수 있습니다. 디코더로 선택한 아키텍처에 따라 교차 어텐션(cross-attention) 레이어가 무작위로 초기화될 수 있습니다. 사전 학습된 인코더와 디코더 체크포인트를 이용해 EncoderDecoderModel을 초기화하려면, 모델을 다운스트림 작업에 대해 미세 조정(fine-tuning)해야 합니다. 이에 대한 자세한 내용은 the Warm-starting-encoder-decoder blog post에 설명되어 있습니다. 이 작업을 위해 EncoderDecoderModel 클래스는 EncoderDecoderModel.from_encoder_decoder_pretrained() 메서드를 제공합니다.
>>> from transformers import EncoderDecoderModel, BertTokenizer
>>> tokenizer = BertTokenizer.from_pretrained("google-bert/bert-base-uncased")
>>> model = EncoderDecoderModel.from_encoder_decoder_pretrained("google-bert/bert-base-uncased", "google-bert/bert-base-uncased")기존 EncoderDecoderModel 체크포인트 불러오기 및 추론하기
EncoderDecoderModel 클래스의 미세 조정(fine-tuned)된 체크포인트를 불러오려면, Transformers의 다른 모델 아키텍처와 마찬가지로 EncoderDecoderModel에서 제공하는 from_pretrained(...)를 사용할 수 있습니다.
추론을 수행하려면 generate 메서드를 활용하여 텍스트를 자동 회귀(autoregressive) 방식으로 생성할 수 있습니다. 이 메서드는 탐욕 디코딩(greedy decoding), 빔 서치(beam search), 다항 샘플링(multinomial sampling) 등 다양한 디코딩 방식을 지원합니다.
>>> from transformers import AutoTokenizer, EncoderDecoderModel
>>> # 미세 조정된 seq2seq 모델과 대응하는 토크나이저 가져오기
>>> model = EncoderDecoderModel.from_pretrained("patrickvonplaten/bert2bert_cnn_daily_mail")
>>> tokenizer = AutoTokenizer.from_pretrained("patrickvonplaten/bert2bert_cnn_daily_mail")
>>> # let's perform inference on a long piece of text
>>> ARTICLE_TO_SUMMARIZE = (
... "PG&E stated it scheduled the blackouts in response to forecasts for high winds "
... "amid dry conditions. The aim is to reduce the risk of wildfires. Nearly 800 thousand customers were "
... "scheduled to be affected by the shutoffs which were expected to last through at least midday tomorrow."
... )
>>> input_ids = tokenizer(ARTICLE_TO_SUMMARIZE, return_tensors="pt").input_ids
>>> # 자기회귀적으로 요약 생성 (기본적으로 그리디 디코딩 사용)
>>> generated_ids = model.generate(input_ids)
>>> generated_text = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
>>> print(generated_text)
nearly 800 thousand customers were affected by the shutoffs. the aim is to reduce the risk of wildfires. nearly 800, 000 customers were expected to be affected by high winds amid dry conditions. pg & e said it scheduled the blackouts to last through at least midday tomorrow.TFEncoderDecoderModel 에 Pytorch 체크포인트 불러오기
TFEncoderDecoderModel.from_pretrained 메서드는 현재 Pytorch 체크포인트를 사용한 모델 초기화를 지원하지 않습니다. 이 메서드에 from_pt=True를 전달하면 예외(exception)가 발생합니다. 특정 인코더-디코더 모델에 대한 Pytorch 체크포인트만 존재하는 경우, 다음과 같은 해결 방법을 사용할 수 있습니다:
>>> # 파이토치 체크포인트에서 로드하는 해결 방법
>>> from transformers import EncoderDecoderModel, TFEncoderDecoderModel
>>> _model = EncoderDecoderModel.from_pretrained("patrickvonplaten/bert2bert-cnn_dailymail-fp16")
>>> _model.encoder.save_pretrained("./encoder")
>>> _model.decoder.save_pretrained("./decoder")
>>> model = TFEncoderDecoderModel.from_encoder_decoder_pretrained(
... "./encoder", "./decoder", encoder_from_pt=True, decoder_from_pt=True
... )
>>> # 이 부분은 특정 모델의 구체적인 세부사항을 복사할 때에만 사용합니다.
>>> model.config = _model.config학습
모델이 생성된 후에는 BART, T5 또는 기타 인코더-디코더 모델과 유사한 방식으로 미세 조정(fine-tuning)할 수 있습니다.
보시다시피, 손실(loss)을 계산하려면 단 2개의 입력만 필요합니다: input_ids(입력 시퀀스를 인코딩한 input_ids)와 labels(목표 시퀀스를 인코딩한 input_ids).
>>> from transformers import BertTokenizer, EncoderDecoderModel
>>> tokenizer = BertTokenizer.from_pretrained("google-bert/bert-base-uncased")
>>> model = EncoderDecoderModel.from_encoder_decoder_pretrained("google-bert/bert-base-uncased", "google-bert/bert-base-uncased")
>>> model.config.decoder_start_token_id = tokenizer.cls_token_id
>>> model.config.pad_token_id = tokenizer.pad_token_id
>>> input_ids = tokenizer(
... "The tower is 324 metres (1,063 ft) tall, about the same height as an 81-storey building, and the tallest structure in Paris. Its base is square, measuring 125 metres (410 ft) on each side.During its construction, the Eiffel Tower surpassed the Washington Monument to become the tallest man-made structure in the world, a title it held for 41 years until the Chrysler Building in New York City was finished in 1930. It was the first structure to reach a height of 300 metres. Due to the addition of a broadcasting aerial at the top of the tower in 1957, it is now taller than the Chrysler Building by 5.2 metres (17 ft).Excluding transmitters, the Eiffel Tower is the second tallest free-standing structure in France after the Millau Viaduct.",
... return_tensors="pt",
... ).input_ids
>>> labels = tokenizer(
... "the eiffel tower surpassed the washington monument to become the tallest structure in the world. it was the first structure to reach a height of 300 metres in paris in 1930. it is now taller than the chrysler building by 5. 2 metres ( 17 ft ) and is the second tallest free - standing structure in paris.",
... return_tensors="pt",
... ).input_ids
>>> # forward 함수가 자동으로 적합한 decoder_input_ids를 생성합니다.
>>> loss = model(input_ids=input_ids, labels=labels).loss훈련에 대한 자세한 내용은 colab 노트북을 참조하세요.
이 모델은 thomwolf가 기여했으며, 이 모델에 대한 TensorFlow 및 Flax 버전은 ydshieh가 기여했습니다.
EncoderDecoderConfig
class transformers.EncoderDecoderConfig
< source >( **kwargs )
Parameters
- kwargs (optional) —
Dictionary of keyword arguments. Notably:
- encoder (PreTrainedConfig, optional) — An instance of a configuration object that defines the encoder config.
- decoder (PreTrainedConfig, optional) — An instance of a configuration object that defines the decoder config.
EncoderDecoderConfig is the configuration class to store the configuration of a EncoderDecoderModel. It is used to instantiate an Encoder Decoder model according to the specified arguments, defining the encoder and decoder configs.
Configuration objects inherit from PreTrainedConfig and can be used to control the model outputs. Read the documentation from PreTrainedConfig for more information.
Examples:
>>> from transformers import BertConfig, EncoderDecoderConfig, EncoderDecoderModel
>>> # Initializing a BERT google-bert/bert-base-uncased style configuration
>>> config_encoder = BertConfig()
>>> config_decoder = BertConfig()
>>> config = EncoderDecoderConfig.from_encoder_decoder_configs(config_encoder, config_decoder)
>>> # Initializing a Bert2Bert model (with random weights) from the google-bert/bert-base-uncased style configurations
>>> model = EncoderDecoderModel(config=config)
>>> # Accessing the model configuration
>>> config_encoder = model.config.encoder
>>> config_decoder = model.config.decoder
>>> # set decoder config to causal lm
>>> config_decoder.is_decoder = True
>>> config_decoder.add_cross_attention = True
>>> # Saving the model, including its configuration
>>> model.save_pretrained("my-model")
>>> # loading model and config from pretrained folder
>>> encoder_decoder_config = EncoderDecoderConfig.from_pretrained("my-model")
>>> model = EncoderDecoderModel.from_pretrained("my-model", config=encoder_decoder_config)from_encoder_decoder_configs
< source >( encoder_config: PreTrainedConfig decoder_config: PreTrainedConfig **kwargs ) → EncoderDecoderConfig
Instantiate a EncoderDecoderConfig (or a derived class) from a pre-trained encoder model configuration and decoder model configuration.
EncoderDecoderModel
class transformers.EncoderDecoderModel
< source >( config: typing.Optional[transformers.configuration_utils.PreTrainedConfig] = None encoder: typing.Optional[transformers.modeling_utils.PreTrainedModel] = None decoder: typing.Optional[transformers.modeling_utils.PreTrainedModel] = None )
Parameters
- config (PreTrainedConfig, optional) — Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.
- encoder (
PreTrainedModel, optional) — The encoder model to use. - decoder (
PreTrainedModel, optional) — The decoder model to use.
The bare Encoder Decoder Model outputting raw hidden-states without any specific head on top.
This model inherits from PreTrainedModel. Check the superclass documentation for the generic methods the library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads etc.)
This model is also a PyTorch torch.nn.Module subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.
forward
< source >( input_ids: typing.Optional[torch.LongTensor] = None attention_mask: typing.Optional[torch.FloatTensor] = None decoder_input_ids: typing.Optional[torch.LongTensor] = None decoder_attention_mask: typing.Optional[torch.BoolTensor] = None encoder_outputs: typing.Optional[tuple[torch.FloatTensor]] = None past_key_values: typing.Optional[transformers.cache_utils.Cache] = None inputs_embeds: typing.Optional[torch.FloatTensor] = None decoder_inputs_embeds: typing.Optional[torch.FloatTensor] = None labels: typing.Optional[torch.LongTensor] = None use_cache: typing.Optional[bool] = None cache_position: typing.Optional[torch.Tensor] = None **kwargs ) → transformers.modeling_outputs.Seq2SeqLMOutput or tuple(torch.FloatTensor)
Parameters
- input_ids (
torch.LongTensorof shape(batch_size, sequence_length), optional) — Indices of input sequence tokens in the vocabulary. Padding will be ignored by default.Indices can be obtained using AutoTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.
- attention_mask (
torch.FloatTensorof shape(batch_size, sequence_length), optional) — Mask to avoid performing attention on padding token indices. Mask values selected in[0, 1]:- 1 for tokens that are not masked,
- 0 for tokens that are masked.
- decoder_input_ids (
torch.LongTensorof shape(batch_size, target_sequence_length), optional) — Indices of decoder input sequence tokens in the vocabulary.Indices can be obtained using PreTrainedTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.
If
past_key_valuesis used, optionally only the lastdecoder_input_idshave to be input (seepast_key_values).For training,
decoder_input_idsare automatically created by the model by shifting thelabelsto the right, replacing -100 by thepad_token_idand prepending them with thedecoder_start_token_id. - decoder_attention_mask (
torch.BoolTensorof shape(batch_size, target_sequence_length), optional) — Default behavior: generate a tensor that ignores pad tokens indecoder_input_ids. Causal mask will also be used by default. - encoder_outputs (
tuple[torch.FloatTensor], optional) — Tuple consists of (last_hidden_state, optional:hidden_states, optional:attentions)last_hidden_stateof shape(batch_size, sequence_length, hidden_size), optional) is a sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention of the decoder. - past_key_values (
~cache_utils.Cache, optional) — Pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention blocks) that can be used to speed up sequential decoding. This typically consists in thepast_key_valuesreturned by the model at a previous stage of decoding, whenuse_cache=Trueorconfig.use_cache=True.Only Cache instance is allowed as input, see our kv cache guide. If no
past_key_valuesare passed, DynamicCache will be initialized by default.The model will output the same cache format that is fed as input.
If
past_key_valuesare used, the user is expected to input only unprocessedinput_ids(those that don’t have their past key value states given to this model) of shape(batch_size, unprocessed_length)instead of allinput_idsof shape(batch_size, sequence_length). - inputs_embeds (
torch.FloatTensorof shape(batch_size, sequence_length, hidden_size), optional) — Optionally, instead of passinginput_idsyou can choose to directly pass an embedded representation. This is useful if you want more control over how to convertinput_idsindices into associated vectors than the model’s internal embedding lookup matrix. - decoder_inputs_embeds (
torch.FloatTensorof shape(batch_size, target_sequence_length, hidden_size), optional) — Optionally, instead of passingdecoder_input_idsyou can choose to directly pass an embedded representation. This is useful if you want more control over how to convertdecoder_input_idsindices into associated vectors than the model’s internal embedding lookup matrix. - labels (
torch.LongTensorof shape(batch_size, sequence_length), optional) — Labels for computing the masked language modeling loss for the decoder. Indices should be in[-100, 0, ..., config.vocab_size](seeinput_idsdocstring) Tokens with indices set to-100are ignored (masked), the loss is only computed for the tokens with labels in[0, ..., config.vocab_size] - use_cache (
bool, optional) — If set toTrue,past_key_valueskey value states are returned and can be used to speed up decoding (seepast_key_values). - cache_position (
torch.Tensorof shape(sequence_length), optional) — Indices depicting the position of the input sequence tokens in the sequence. Contrarily toposition_ids, this tensor is not affected by padding. It is used to update the cache in the correct position and to infer the complete sequence length.
Returns
transformers.modeling_outputs.Seq2SeqLMOutput or tuple(torch.FloatTensor)
A transformers.modeling_outputs.Seq2SeqLMOutput or a tuple of
torch.FloatTensor (if return_dict=False is passed or when config.return_dict=False) comprising various
elements depending on the configuration (EncoderDecoderConfig) and inputs.
-
loss (
torch.FloatTensorof shape(1,), optional, returned whenlabelsis provided) — Language modeling loss. -
logits (
torch.FloatTensorof shape(batch_size, sequence_length, config.vocab_size)) — Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax). -
past_key_values (
EncoderDecoderCache, optional, returned whenuse_cache=Trueis passed or whenconfig.use_cache=True) — It is a EncoderDecoderCache instance. For more details, see our kv cache guide.Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention blocks) that can be used (see
past_key_valuesinput) to speed up sequential decoding. -
decoder_hidden_states (
tuple(torch.FloatTensor), optional, returned whenoutput_hidden_states=Trueis passed or whenconfig.output_hidden_states=True) — Tuple oftorch.FloatTensor(one for the output of the embeddings, if the model has an embedding layer, + one for the output of each layer) of shape(batch_size, sequence_length, hidden_size).Hidden-states of the decoder at the output of each layer plus the initial embedding outputs.
-
decoder_attentions (
tuple(torch.FloatTensor), optional, returned whenoutput_attentions=Trueis passed or whenconfig.output_attentions=True) — Tuple oftorch.FloatTensor(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length).Attentions weights of the decoder, after the attention softmax, used to compute the weighted average in the self-attention heads.
-
cross_attentions (
tuple(torch.FloatTensor), optional, returned whenoutput_attentions=Trueis passed or whenconfig.output_attentions=True) — Tuple oftorch.FloatTensor(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length).Attentions weights of the decoder’s cross-attention layer, after the attention softmax, used to compute the weighted average in the cross-attention heads.
-
encoder_last_hidden_state (
torch.FloatTensorof shape(batch_size, sequence_length, hidden_size), optional) — Sequence of hidden-states at the output of the last layer of the encoder of the model. -
encoder_hidden_states (
tuple(torch.FloatTensor), optional, returned whenoutput_hidden_states=Trueis passed or whenconfig.output_hidden_states=True) — Tuple oftorch.FloatTensor(one for the output of the embeddings, if the model has an embedding layer, + one for the output of each layer) of shape(batch_size, sequence_length, hidden_size).Hidden-states of the encoder at the output of each layer plus the initial embedding outputs.
-
encoder_attentions (
tuple(torch.FloatTensor), optional, returned whenoutput_attentions=Trueis passed or whenconfig.output_attentions=True) — Tuple oftorch.FloatTensor(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length).Attentions weights of the encoder, after the attention softmax, used to compute the weighted average in the self-attention heads.
The EncoderDecoderModel forward method, overrides the __call__ special method.
Although the recipe for forward pass needs to be defined within this function, one should call the
Moduleinstance afterwards instead of this since the former takes care of running the pre and post processing steps while the latter silently ignores them.
Examples:
>>> from transformers import EncoderDecoderModel, BertTokenizer
>>> import torch
>>> tokenizer = BertTokenizer.from_pretrained("google-bert/bert-base-uncased")
>>> model = EncoderDecoderModel.from_encoder_decoder_pretrained(
... "google-bert/bert-base-uncased", "google-bert/bert-base-uncased"
... ) # initialize Bert2Bert from pre-trained checkpoints
>>> # training
>>> model.config.decoder_start_token_id = tokenizer.cls_token_id
>>> model.config.pad_token_id = tokenizer.pad_token_id
>>> model.config.vocab_size = model.config.decoder.vocab_size
>>> input_ids = tokenizer("This is a really long text", return_tensors="pt").input_ids
>>> labels = tokenizer("This is the corresponding summary", return_tensors="pt").input_ids
>>> outputs = model(input_ids=input_ids, labels=labels)
>>> loss, logits = outputs.loss, outputs.logits
>>> # save and load from pretrained
>>> model.save_pretrained("bert2bert")
>>> model = EncoderDecoderModel.from_pretrained("bert2bert")
>>> # generation
>>> generated = model.generate(input_ids)from_encoder_decoder_pretrained
< source >( encoder_pretrained_model_name_or_path: typing.Optional[str] = None decoder_pretrained_model_name_or_path: typing.Optional[str] = None *model_args **kwargs )
Parameters
- encoder_pretrained_model_name_or_path (
str, optional) — Information necessary to initiate the encoder. Can be either:- A string, the model id of a pretrained model hosted inside a model repo on huggingface.co.
- A path to a directory containing model weights saved using
save_pretrained(), e.g.,
./my_model_directory/.
- decoder_pretrained_model_name_or_path (
str, optional, defaults toNone) — Information necessary to initiate the decoder. Can be either:- A string, the model id of a pretrained model hosted inside a model repo on huggingface.co.
- A path to a directory containing model weights saved using
save_pretrained(), e.g.,
./my_model_directory/.
- model_args (remaining positional arguments, optional) —
All remaining positional arguments will be passed to the underlying model’s
__init__method. - kwargs (remaining dictionary of keyword arguments, optional) —
Can be used to update the configuration object (after it being loaded) and initiate the model (e.g.,
output_attentions=True).- To update the encoder configuration, use the prefix encoder_ for each configuration parameter.
- To update the decoder configuration, use the prefix decoder_ for each configuration parameter.
- To update the parent model configuration, do not use a prefix for each configuration parameter.
Behaves differently depending on whether a
configis provided or automatically loaded.
Instantiate an encoder and a decoder from one or two base classes of the library from pretrained model checkpoints.
The model is set in evaluation mode by default using model.eval() (Dropout modules are deactivated). To train
the model, you need to first set it back in training mode with model.train().
Example:
>>> from transformers import EncoderDecoderModel
>>> # initialize a bert2bert from two pretrained BERT models. Note that the cross-attention layers will be randomly initialized
>>> model = EncoderDecoderModel.from_encoder_decoder_pretrained("google-bert/bert-base-uncased", "google-bert/bert-base-uncased")
>>> # saving model after fine-tuning
>>> model.save_pretrained("./bert2bert")
>>> # load fine-tuned model
>>> model = EncoderDecoderModel.from_pretrained("./bert2bert")