flair.embeddings.transformer.TransformerEmbeddings#

class flair.embeddings.transformer.TransformerEmbeddings(model='bert-base-uncased', fine_tune=True, layers='-1', layer_mean=True, subtoken_pooling='first', cls_pooling='cls', is_token_embedding=True, is_document_embedding=True, allow_long_sentences=False, use_context=False, respect_document_boundaries=True, context_dropout=0.5, saved_config=None, tokenizer_data=None, feature_extractor_data=None, name=None, force_max_length=False, needs_manual_ocr=None, use_context_separator=True, transformers_tokenizer_kwargs={}, transformers_config_kwargs={}, transformers_model_kwargs={}, peft_config=None, peft_gradient_checkpointing_kwargs={}, **kwargs)View on GitHub#

Bases: TransformerBaseEmbeddings

__init__(model='bert-base-uncased', fine_tune=True, layers='-1', layer_mean=True, subtoken_pooling='first', cls_pooling='cls', is_token_embedding=True, is_document_embedding=True, allow_long_sentences=False, use_context=False, respect_document_boundaries=True, context_dropout=0.5, saved_config=None, tokenizer_data=None, feature_extractor_data=None, name=None, force_max_length=False, needs_manual_ocr=None, use_context_separator=True, transformers_tokenizer_kwargs={}, transformers_config_kwargs={}, transformers_model_kwargs={}, peft_config=None, peft_gradient_checkpointing_kwargs={}, **kwargs)View on GitHub#

Instantiate transformers embeddings.

Allows using transformers as TokenEmbeddings and DocumentEmbeddings or both.

Parameters:
  • model (str) – name of transformer model (see huggingface hub for options)

  • fine_tune (bool) – If True, the weights of the transformers embedding will be updated during training.

  • layers (str) – Specify which layers should be extracted for the embeddings. Expects either “all” to extract all layers or a comma separated list of indices (e.g. “-1,-2,-3,-4” for the last 4 layers)

  • layer_mean (bool) – If True, the extracted layers will be averaged. Otherwise, they will be concatenated.

  • subtoken_pooling (Literal['first', 'last', 'first_last', 'mean']) – Specify how multiple sub-tokens will be aggregated for a token-embedding.

  • cls_pooling (Literal['cls', 'max', 'mean']) – Specify how the document-embeddings will be extracted.

  • is_token_embedding (bool) – If True, this embeddings can be handled as token-embeddings.

  • is_document_embedding (bool) – If True, this embeddings can be handled document-embeddings.

  • allow_long_sentences (bool) – If True, too long sentences will be patched and strided and afterwards combined.

  • use_context (Union[bool, int]) – If True, predicting multiple sentences at once, will use the previous and next sentences for context.

  • respect_document_boundaries (bool) – If True, the context calculation will stop if a sentence represents a context boundary.

  • context_dropout (float) – Integer percentage (0-100) to specify how often the context won’t be used during training.

  • saved_config (Optional[PretrainedConfig]) – Pretrained config used when loading embeddings. Always use None.

  • tokenizer_data (Optional[BytesIO]) – Tokenizer data used when loading embeddings. Always use None.

  • feature_extractor_data (Optional[BytesIO]) – Feature extractor data used when loading embeddings. Always use None.

  • name (Optional[str]) – The name for the embeddings. Per default the name will be used from the used transformers model.

  • force_max_length (bool) – If True, the tokenizer will always pad the sequences to maximum length.

  • needs_manual_ocr (Optional[bool]) – If True, bounding boxes will be calculated manually. This is used for models like layoutlm where the tokenizer doesn’t compute the bounding boxes itself.

  • use_context_separator (bool) – If True, the embedding will hold an additional token to allow the model to distingulish between context and prediction.

  • transformers_tokenizer_kwargs (dict[str, Any]) – Further values forwarded to the initialization of the transformers tokenizer

  • transformers_config_kwargs (dict[str, Any]) – Further values forwarded to the initialization of the transformers config

  • transformers_model_kwargs (dict[str, Any]) – Further values forwarded to the initialization of the transformers model

  • peft_config – If set, the model will be trained using adapters and optionally QLoRA. Should be of type “PeftConfig” or subtype

  • peft_gradient_checkpointing_kwargs (Optional[dict[str, Any]]) – Further values used when preparing the model for kbit training. Only used if peft_config is set.

  • **kwargs – Further values forwarded to the transformers config

Methods

__init__([model, fine_tune, layers, ...])

Instantiate transformers embeddings.

add_module(name, module)

Add a child module to the current module.

apply(fn)

Apply fn recursively to every submodule (as returned by .children()) as well as self.

bfloat16()

Casts all floating point parameters and buffers to bfloat16 datatype.

buffers([recurse])

Return an iterator over module buffers.

children()

Return an iterator over immediate children modules.

compile(*args, **kwargs)

Compile this Module's forward using torch.compile().

cpu()

Move all model parameters and buffers to the CPU.

create_from_state(**state)

cuda([device])

Move all model parameters and buffers to the GPU.

double()

Casts all floating point parameters and buffers to double datatype.

embed(data_points)

Add embeddings to all words in a list of sentences.

eval()

Set the module in evaluation mode.

export_onnx(path, example_sentences, **kwargs)

Export TransformerEmbeddings to OnnxFormat.

extra_repr()

Set the extra representation of the module.

float()

Casts all floating point parameters and buffers to float datatype.

forward(input_ids[, sub_token_lengths, ...])

Define the computation performed at every call.

from_params(params)

get_buffer(target)

Return the buffer given by target if it exists, otherwise throw an error.

get_extra_state()

Return any extra state to include in the module's state_dict.

get_instance_parameters(locals)

get_names()

Returns a list of embedding names.

get_parameter(target)

Return the parameter given by target if it exists, otherwise throw an error.

get_submodule(target)

Return the submodule given by target if it exists, otherwise throw an error.

half()

Casts all floating point parameters and buffers to half datatype.

ipu([device])

Move all model parameters and buffers to the IPU.

load_embedding(params)

load_state_dict(state_dict[, strict, assign])

Copy parameters and buffers from state_dict into this module and its descendants.

modules()

Return an iterator over all modules in the network.

mtia([device])

Move all model parameters and buffers to the MTIA.

named_buffers([prefix, recurse, ...])

Return an iterator over module buffers, yielding both the name of the buffer as well as the buffer itself.

named_children()

Return an iterator over immediate children modules, yielding both the name of the module as well as the module itself.

named_modules([memo, prefix, remove_duplicate])

Return an iterator over all modules in the network, yielding both the name of the module as well as the module itself.

named_parameters([prefix, recurse, ...])

Return an iterator over module parameters, yielding both the name of the parameter as well as the parameter itself.

parameters([recurse])

Return an iterator over module parameters.

prepare_tensors(sentences[, device])

register_backward_hook(hook)

Register a backward hook on the module.

register_buffer(name, tensor[, persistent])

Add a buffer to the module.

register_forward_hook(hook, *[, prepend, ...])

Register a forward hook on the module.

register_forward_pre_hook(hook, *[, ...])

Register a forward pre-hook on the module.

register_full_backward_hook(hook[, prepend])

Register a backward hook on the module.

register_full_backward_pre_hook(hook[, prepend])

Register a backward pre-hook on the module.

register_load_state_dict_post_hook(hook)

Register a post-hook to be run after module's load_state_dict() is called.

register_load_state_dict_pre_hook(hook)

Register a pre-hook to be run before module's load_state_dict() is called.

register_module(name, module)

Alias for add_module().

register_parameter(name, param)

Add a parameter to the module.

register_state_dict_post_hook(hook)

Register a post-hook for the state_dict() method.

register_state_dict_pre_hook(hook)

Register a pre-hook for the state_dict() method.

requires_grad_([requires_grad])

Change if autograd should record operations on parameters in this module.

save_embeddings([use_state_dict])

set_extra_state(state)

Set extra state contained in the loaded state_dict.

set_submodule(target, module)

Set the submodule given by target if it exists, otherwise throw an error.

share_memory()

See torch.Tensor.share_memory_().

state_dict(*args[, destination, prefix, ...])

Return a dictionary containing references to the whole state of the module.

to(*args, **kwargs)

Move and/or cast the parameters and buffers.

to_args()

to_empty(*, device[, recurse])

Move the parameters and buffers to the specified device without copying storage.

to_params()

train([mode])

Set the module in training mode.

type(dst_type)

Casts all parameters and buffers to dst_type.

xpu([device])

Move all model parameters and buffers to the XPU.

zero_grad([set_to_none])

Reset gradients of all model parameters.

Attributes

T_destination

call_super_init

dump_patches

embedding_length

Returns the length of the embedding vector.

embedding_type

embeddings_name

training

onnx_clsView on GitHub#

alias of TransformerOnnxEmbeddings

embeddings_name: str = 'TransformerEmbeddings'#
property embedding_length: int#

Returns the length of the embedding vector.

property embedding_type: str#
classmethod from_params(params)View on GitHub#
to_params()View on GitHub#
forward(input_ids, sub_token_lengths=None, token_lengths=None, attention_mask=None, overflow_to_sample_mapping=None, word_ids=None, langs=None, bbox=None, pixel_values=None)View on GitHub#

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

export_onnx(path, example_sentences, **kwargs)View on GitHub#

Export TransformerEmbeddings to OnnxFormat.

Parameters:
  • path (Union[str, Path]) – the path to save the embeddings. Notice that the embeddings are stored as external file, hence it matters if the path is an absolue path or a relative one.

  • example_sentences (list[Sentence]) – a list of sentences that will be used for tracing. It is recommended to take 2-4 sentences with some variation.

  • **kwargs – the parameters passed to TransformerOnnxEmbeddings.export_from_embedding()

Return type:

TransformerOnnxEmbeddings