MEGA
Last updated
Last updated
The MEGA model was proposed in by Xuezhe Ma, Chunting Zhou, Xiang Kong, Junxian He, Liangke Gui, Graham Neubig, Jonathan May, and Luke Zettlemoyer. MEGA proposes a new approach to self-attention with each encoder layer having a multi-headed exponential moving average in addition to a single head of standard dot-product attention, giving the attention mechanism stronger positional biases. This allows MEGA to perform competitively to Transformers on standard benchmarks including LRA while also having significantly fewer parameters. MEGA’s compute efficiency allows it to scale to very long sequences, making it an attractive option for long-document NLP tasks.
The abstract from the paper is the following:
The design choices in the Transformer attention mechanism, including weak inductive bias and quadratic computational complexity, have limited its application for modeling long sequences. In this paper, we introduce Mega, a simple, theoretically grounded, single-head gated attention mechanism equipped with (exponential) moving average to incorporate inductive bias of position-aware local dependencies into the position-agnostic attention mechanism. We further propose a variant of Mega that offers linear time and space complexity yet yields only minimal quality loss, by efficiently splitting the whole sequence into multiple chunks with fixed length. Extensive experiments on a wide range of sequence modeling benchmarks, including the Long Range Arena, neural machine translation, auto-regressive language modeling, and image and speech classification, show that Mega achieves significant improvements over other sequence models, including variants of Transformers and recent state space models.
Tips:
MEGA can perform quite well with relatively few parameters. See Appendix D in the MEGA paper for examples of architectural specs which perform well in various settings. If using MEGA as a decoder, be sure to set bidirectional=False to avoid errors with default bidirectional.
Mega-chunk is a variant of mega that reduces time and spaces complexity from quadratic to linear. Utilize chunking with MegaConfig.use_chunking and control chunk size with MegaConfig.chunk_size
This model was contributed by . The original code can be found .
Implementation Notes:
The original implementation of MEGA had an inconsistent expectation of attention masks for padding and causal self-attention between the softmax attention and Laplace/squared ReLU method. This implementation addresses that inconsistency.
The original implementation did not include token type embeddings; this implementation adds support for these, with the option controlled by MegaConfig.add_token_type_embeddings
( vocab_size = 30522hidden_size = 128num_hidden_layers = 4intermediate_size = 256ema_projection_size = 16bidirectional = Trueshared_representation_size = 64use_chunking = Falsechunk_size = -1truncation = Nonenormalize_before_mega = Truenormalization_type = 'scalenorm'norm_affine = Trueactivation = 'silu'attention_activation = 'softmax'dropout_prob = 0.1hidden_dropout_prob = 0.1attention_probs_dropout_prob = 0.1use_feature_dropout = Falseuse_normalized_ffn = Truenffn_hidden_size = 256normalize_before_ffn = Truenffn_activation_dropout_prob = 0.1max_positions = 2048add_token_type_embeddings = Falsetype_vocab_size = 2initializer_range = 0.02ema_delta_alpha_range = 0.2ema_beta_range = 0.02ema_gamma_omega_range = 1.0pad_token_id = 1bos_token_id = 0eos_token_id = 2relative_positional_bias = 'rotary'classifier_dropout = Noneuse_cache = Trueadd_lm_hidden_dense_layer = True**kwargs )
Parameters
hidden_size (int, optional, defaults to 128) — Dimensionality of the encoder layers and the pooler layer.
num_hidden_layers (int, optional, defaults to 4) — Number of hidden layers in the Mega encoder.
intermediate_size (int, optional, defaults to 256) — Dimensionality of the hidden size (self-attention value projection) within the Mega encoder
ema_projection_size (int, optional, defaults to 16) — Dimensionality of the MegaMultiDimensionDampedEma
bidirectional (bool, optional, defaults to True) — Whether the MegaMultiDimensionDampedEma used in Mega’s self-attention should work bidirectionally (True) or unidirectionally (False). Bidirectional EMA is incompatible with causal decoding, so this should be False if you intend to use the model as a decoder.
shared_representation_size (int, optional, defaults to 64) — Dimensionality of the linear projection for shared representation of self-attention queries and keys
use_chunking (bool, optional, defaults to False) — Whether to chunk inputs for linear self-attention complexity (described as Mega-chunk in the paper)
chunk_size (int, optional, defaults to -1) — If use_chunking is set to True, determines the size of the chunks to apply to the input sequence. If chunking is used, input sequences must be padded to a multiple of chunk_size
truncation (int, optional) — If specified, the sequence length for which to truncate MegaMultiDimensionDampedEma
normalize_before_mega (bool, optional, defaults to True) — Whether to normalize before (True) or after (False) passing through Mega encoder blocks
normalization_type (str, optional, defaults to "scalenorm") — Type of normalization to use in Mega encoder blocks. Choose one of "scalenorm", "layernorm", "rmsnorm", "batchnorm", or "syncbatchnorm" (GPU required for syncbatchnorm)
norm_affine (bool, optional, defaults to True) — If True, applies a parameterized affine transformation to inputs during normalization
activation (str, optional, defaults to "silu") — Activation function to apply within Mega encoder blocks. Choose one of "silu", "relu", "linear", "gelu", or "gelu_accurate"
attention_activation (str, optional, defaults to "softmax") — Activation function to apply for single-headed self-attention (a la Transformer). Choose one of "softmax", "laplace", or "relu2"
dropout_prob (float, optional, defaults to 0.1) — The dropout probability for EMA self-attention
hidden_dropout_prob (float, optional, defaults to 0.1) — The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.
attention_probs_dropout_prob (float, optional, defaults to 0.1) — The dropout ratio for the attention probabilities.
use_feature_dropout (bool, optional, defaults to False) — Whether to use feature-based (True) or standard dropout (False)
use_normalized_ffn (bool, optional, defaults to True) — Whether to use the normalized feed-forward sub-layer in Mega blocks (True) or pass Mega encoder output as-is (False)
nffn_hidden_size (int, optional, defaults to 256) — If using the normalized feed-forward network (NFFN) layer within Mega (use_normalized_ffn = True), this is the hidden size of the NFFN
normalize_before_ffn (bool, optional, defaults to True) — Whether to normalize before (True) or after (False) the feed-forward portion of NFFN
nffn_activation_dropout_prob (float, optional, defaults to 0.1) — The dropout ratio for the NFFN component.
max_positions (int, optional, defaults to 2048) — The maximum sequence length to use for positional representations. For "simple" relative positional bias, this is a hard limit on input length; "rotary" relative positional bias will extrapolate to longer sequences
add_token_type_embeddings (bool, optional, defaults to True) — Whether to account for token types in embeddings. Left as optional to maintain compatibility with original implementation while adding support for token types.
initializer_range (float, optional, defaults to 0.02) — The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
ema_delta_alpha_range (float, optional, defaults to 0.2) — The standard deviation for initializing the delta (damping factor) and alpha (decay factor) parameters in MegaMultiDimensionDampedEma.
ema_beta_range (float, optional, defaults to 0.02) — The standard deviation for initializing the beta parameter (expansion matrix) in MegaMultiDimensionDampedEma.
ema_gamma_omega_range (float, optional, defaults to 1.0) — The standard deviation for initializing the gamma (projection matrix) and omega (residual weight) parameters in MultiDimensionEMA.
relative_positional_bias (str, optional, defaults to "rotary") — Type of relative positional encoding. Choose one of "rotary" or "simple". If "simple" is selected, max_positions is used as a limit on input size, while "rotary" extrapolates beyond max_positions.
is_decoder (bool, optional, defaults to False) — Whether the model is used as a decoder or not. If False, the model is used as an encoder.
use_cache (bool, optional, defaults to True) — Whether or not the model should return the last key/values attentions (not used by all models). Only relevant if config.is_decoder=True.
classifier_dropout (float, optional) — The dropout ratio for the classification head.
add_lm_hidden_dense_layer (bool, optional, defaults to True) — Whether to include a hidden layer for projection between encoder outputs and LM heads (True) or pass hidden states directly to LM head (False). Remains optional for compatibility with original implementation
Examples:
Copied
( config: MegaConfigadd_pooling_layer = True )
Parameters
The bare MEGA Model transformer outputting raw hidden-states without any specific head on top.
The model can behave as an encoder (with only self-attention) as well as a decoder, in which case a layer of cross-attention is added after self-attention, following the architecture described in Mega: Moving Average Equipped Gated Attention_ by Xuezhe Ma, Chunting Zhou, Xiang Kong, Junxian He, Liangke Gui, Graham Neubig, Jonathan May, and Luke Zettlemoyer
To behave as a decoder the model needs to be initialized with the is_decoder argument of the configuration set to True and bidirectional set to False. To be used in a Seq2Seq model, the model needs to initialized with both is_decoder=True and bidirectional=False argument as well as add_cross_attention set to True; an encoder_hidden_states is then expected as an input to the forward pass.
forward
Parameters
input_ids (torch.LongTensor of shape (batch_size, sequence_length)) — Indices of input sequence tokens in the vocabulary.
attention_mask (torch.FloatTensor of 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.
token_type_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) — Segment token indices to indicate first and second portions of the inputs. Indices are selected in [0,1]:
0 corresponds to a sentence A token,
1 corresponds to a sentence B token. This parameter can only be used when the model is initialized with add_token_type_embeddings parameter set to True. All the value in this tensor should be always < config.type_vocab_size.
inputs_embeds (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size), optional) — Optionally, instead of passing input_ids you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert input_ids indices into associated vectors than the model’s internal embedding lookup matrix.
output_attentions (bool, optional) — Whether or not to return the attentions tensors of all attention layers. See attentions under returned tensors for more detail.
output_hidden_states (bool, optional) — Whether or not to return the hidden states of all layers. See hidden_states under returned tensors for more detail.
encoder_hidden_states (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size), optional) — Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if the model is configured as a decoder.
encoder_attention_mask (torch.FloatTensor of shape (batch_size, sequence_length), optional) — Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in the cross-attention if the model is configured as a decoder. Mask values selected in [0, 1]:
1 for tokens that are not masked,
0 for tokens that are masked.
past_key_values (tuple(tuple(torch.FloatTensor)) of length config.n_layers with each tuple having 4 tensors of shape (batch_size, num_heads, sequence_length - 1, embed_size_per_head)) — Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up decoding.
If past_key_values are used, the user can optionally input only the last decoder_input_ids (those that don’t have their past key value states given to this model) of shape (batch_size, 1) instead of all decoder_input_ids of shape (batch_size, sequence_length).
use_cache (bool, optional) — If set to True, past_key_values key value states are returned and can be used to speed up decoding (see past_key_values).
Returns
last_hidden_state (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size)) — Sequence of hidden-states at the output of the last layer of the model.
pooler_output (torch.FloatTensor of shape (batch_size, hidden_size)) — Last layer hidden-state of the first token of the sequence (classification token) after further processing through the layers used for the auxiliary pretraining task. E.g. for BERT-family of models, this returns the classification token after processing through a linear layer and a tanh activation function. The linear layer weights are trained from the next sentence prediction (classification) objective during pretraining.
hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) — Tuple of torch.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 model at the output of each layer plus the optional initial embedding outputs.
attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) — Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).
Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
cross_attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True and config.add_cross_attention=True is passed or when config.output_attentions=True) — Tuple of torch.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.
past_key_values (tuple(tuple(torch.FloatTensor)), optional, returned when use_cache=True is passed or when config.use_cache=True) — Tuple of tuple(torch.FloatTensor) of length config.n_layers, with each tuple having 2 tensors of shape (batch_size, num_heads, sequence_length, embed_size_per_head)) and optionally if config.is_encoder_decoder=True 2 additional tensors of shape (batch_size, num_heads, encoder_sequence_length, embed_size_per_head).
Contains pre-computed hidden-states (key and values in the self-attention blocks and optionally if config.is_encoder_decoder=True in the cross-attention blocks) that can be used (see past_key_values input) to speed up sequential decoding.
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 pre and post processing steps while the latter silently ignores them.
Example:
Copied
( config: MegaConfig )
Parameters
MEGA Model with a language modeling head on top for CLM fine-tuning.
forward
Parameters
input_ids (torch.LongTensor of shape (batch_size, sequence_length)) — Indices of input sequence tokens in the vocabulary.
attention_mask (torch.FloatTensor of 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.
token_type_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) — Segment token indices to indicate first and second portions of the inputs. Indices are selected in [0,1]:
0 corresponds to a sentence A token,
1 corresponds to a sentence B token. This parameter can only be used when the model is initialized with add_token_type_embeddings parameter set to True. All the value in this tensor should be always < config.type_vocab_size.
inputs_embeds (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size), optional) — Optionally, instead of passing input_ids you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert input_ids indices into associated vectors than the model’s internal embedding lookup matrix.
output_attentions (bool, optional) — Whether or not to return the attentions tensors of all attention layers. See attentions under returned tensors for more detail.
output_hidden_states (bool, optional) — Whether or not to return the hidden states of all layers. See hidden_states under returned tensors for more detail.
encoder_hidden_states (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size), optional) — Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if the model is configured as a decoder.
encoder_attention_mask (torch.FloatTensor of shape (batch_size, sequence_length), optional) — Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in the cross-attention if the model is configured as a decoder. Mask values selected in [0, 1]:
1 for tokens that are not masked,
0 for tokens that are masked.
labels (torch.LongTensor of shape (batch_size, sequence_length), optional) — Labels for computing the left-to-right language modeling loss (next word prediction). Indices should be in [-100, 0, ..., config.vocab_size] (see input_ids docstring) Tokens with indices set to -100 are ignored (masked), the loss is only computed for the tokens with labels in [0, ..., config.vocab_size]
past_key_values (tuple(tuple(torch.FloatTensor)) of length config.n_layers with each tuple having 4 tensors of shape (batch_size, num_heads, sequence_length - 1, embed_size_per_head)) — Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up decoding.
If past_key_values are used, the user can optionally input only the last decoder_input_ids (those that don’t have their past key value states given to this model) of shape (batch_size, 1) instead of all decoder_input_ids of shape (batch_size, sequence_length).
use_cache (bool, optional) — If set to True, past_key_values key value states are returned and can be used to speed up decoding (see past_key_values).
Returns
loss (torch.FloatTensor of shape (1,), optional, returned when labels is provided) — Language modeling loss (for next-token prediction).
logits (torch.FloatTensor of shape (batch_size, sequence_length, config.vocab_size)) — Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) — Tuple of torch.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 model at the output of each layer plus the optional initial embedding outputs.
attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) — Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).
Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
cross_attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) — Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).
Cross attentions weights after the attention softmax, used to compute the weighted average in the cross-attention heads.
past_key_values (tuple(tuple(torch.FloatTensor)), optional, returned when use_cache=True is passed or when config.use_cache=True) — Tuple of torch.FloatTensor tuples of length config.n_layers, with each tuple containing the cached key, value states of the self-attention and the cross-attention layers if model is used in encoder-decoder setting. Only relevant if config.is_decoder = True.
Contains pre-computed hidden-states (key and values in the attention blocks) that can be used (see past_key_values input) to speed up sequential decoding.
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 pre and post processing steps while the latter silently ignores them.
Example:
Copied
( config: MegaConfig )
Parameters
MEGA Model with a language modeling head on top.
forward
Parameters
input_ids (torch.LongTensor of shape (batch_size, sequence_length)) — Indices of input sequence tokens in the vocabulary.
attention_mask (torch.FloatTensor of 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.
token_type_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) — Segment token indices to indicate first and second portions of the inputs. Indices are selected in [0,1]:
0 corresponds to a sentence A token,
1 corresponds to a sentence B token. This parameter can only be used when the model is initialized with add_token_type_embeddings parameter set to True. All the value in this tensor should be always < config.type_vocab_size.
inputs_embeds (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size), optional) — Optionally, instead of passing input_ids you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert input_ids indices into associated vectors than the model’s internal embedding lookup matrix.
output_attentions (bool, optional) — Whether or not to return the attentions tensors of all attention layers. See attentions under returned tensors for more detail.
output_hidden_states (bool, optional) — Whether or not to return the hidden states of all layers. See hidden_states under returned tensors for more detail.
labels (torch.LongTensor of shape (batch_size, sequence_length), optional) — Labels for computing the masked language modeling loss. Indices should be in [-100, 0, ..., config.vocab_size] (see input_ids docstring) Tokens with indices set to -100 are ignored (masked), the loss is only computed for the tokens with labels in [0, ..., config.vocab_size]
kwargs (Dict[str, any], optional, defaults to {}) — Used to hide legacy arguments that have been deprecated.
Returns
loss (torch.FloatTensor of shape (1,), optional, returned when labels is provided) — Masked language modeling (MLM) loss.
logits (torch.FloatTensor of shape (batch_size, sequence_length, config.vocab_size)) — Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) — Tuple of torch.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 model at the output of each layer plus the optional initial embedding outputs.
attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) — Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).
Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
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 pre and post processing steps while the latter silently ignores them.
Example:
Copied
( config )
Parameters
MEGA Model transformer with a sequence classification/regression head on top (a linear layer on top of the pooled output) e.g. for GLUE tasks.
forward
Parameters
input_ids (torch.LongTensor of shape (batch_size, sequence_length)) — Indices of input sequence tokens in the vocabulary.
attention_mask (torch.FloatTensor of 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.
token_type_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) — Segment token indices to indicate first and second portions of the inputs. Indices are selected in [0,1]:
0 corresponds to a sentence A token,
1 corresponds to a sentence B token. This parameter can only be used when the model is initialized with add_token_type_embeddings parameter set to True. All the value in this tensor should be always < config.type_vocab_size.
inputs_embeds (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size), optional) — Optionally, instead of passing input_ids you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert input_ids indices into associated vectors than the model’s internal embedding lookup matrix.
output_attentions (bool, optional) — Whether or not to return the attentions tensors of all attention layers. See attentions under returned tensors for more detail.
output_hidden_states (bool, optional) — Whether or not to return the hidden states of all layers. See hidden_states under returned tensors for more detail.
labels (torch.LongTensor of shape (batch_size,), optional) — Labels for computing the sequence classification/regression loss. Indices should be in [0, ..., config.num_labels - 1]. If config.num_labels == 1 a regression loss is computed (Mean-Square loss), If config.num_labels > 1 a classification loss is computed (Cross-Entropy).
Returns
loss (torch.FloatTensor of shape (1,), optional, returned when labels is provided) — Classification (or regression if config.num_labels==1) loss.
logits (torch.FloatTensor of shape (batch_size, config.num_labels)) — Classification (or regression if config.num_labels==1) scores (before SoftMax).
hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) — Tuple of torch.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 model at the output of each layer plus the optional initial embedding outputs.
attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) — Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).
Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
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 pre and post processing steps while the latter silently ignores them.
Example of single-label classification:
Copied
Example of multi-label classification:
Copied
( config )
Parameters
MEGA Model with a multiple choice classification head on top (a linear layer on top of the pooled output and a softmax) e.g. for RocStories/SWAG tasks.
forward
Parameters
input_ids (torch.LongTensor of shape (batch_size, num_choices, sequence_length)) — Indices of input sequence tokens in the vocabulary.
attention_mask (torch.FloatTensor of shape (batch_size, num_choices, 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.
token_type_ids (torch.LongTensor of shape (batch_size, num_choices, sequence_length), optional) — Segment token indices to indicate first and second portions of the inputs. Indices are selected in [0,1]:
0 corresponds to a sentence A token,
1 corresponds to a sentence B token. This parameter can only be used when the model is initialized with add_token_type_embeddings parameter set to True. All the value in this tensor should be always < config.type_vocab_size.
inputs_embeds (torch.FloatTensor of shape (batch_size, num_choices, sequence_length, hidden_size), optional) — Optionally, instead of passing input_ids you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert input_ids indices into associated vectors than the model’s internal embedding lookup matrix.
output_attentions (bool, optional) — Whether or not to return the attentions tensors of all attention layers. See attentions under returned tensors for more detail.
output_hidden_states (bool, optional) — Whether or not to return the hidden states of all layers. See hidden_states under returned tensors for more detail.
labels (torch.LongTensor of shape (batch_size,), optional) — Labels for computing the multiple choice classification loss. Indices should be in [0, ..., num_choices-1] where num_choices is the size of the second dimension of the input tensors. (See input_ids above)
Returns
loss (torch.FloatTensor of shape (1,), optional, returned when labels is provided) — Classification loss.
logits (torch.FloatTensor of shape (batch_size, num_choices)) — num_choices is the second dimension of the input tensors. (see input_ids above).
Classification scores (before SoftMax).
hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) — Tuple of torch.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 model at the output of each layer plus the optional initial embedding outputs.
attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) — Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).
Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
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 pre and post processing steps while the latter silently ignores them.
Example:
Copied
( config )
Parameters
MEGA Model with a token classification head on top (a linear layer on top of the hidden-states output) e.g. for Named-Entity-Recognition (NER) tasks.
forward
Parameters
input_ids (torch.LongTensor of shape (batch_size, sequence_length)) — Indices of input sequence tokens in the vocabulary.
attention_mask (torch.FloatTensor of 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.
token_type_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) — Segment token indices to indicate first and second portions of the inputs. Indices are selected in [0,1]:
0 corresponds to a sentence A token,
1 corresponds to a sentence B token. This parameter can only be used when the model is initialized with add_token_type_embeddings parameter set to True. All the value in this tensor should be always < config.type_vocab_size.
inputs_embeds (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size), optional) — Optionally, instead of passing input_ids you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert input_ids indices into associated vectors than the model’s internal embedding lookup matrix.
output_attentions (bool, optional) — Whether or not to return the attentions tensors of all attention layers. See attentions under returned tensors for more detail.
output_hidden_states (bool, optional) — Whether or not to return the hidden states of all layers. See hidden_states under returned tensors for more detail.
labels (torch.LongTensor of shape (batch_size, sequence_length), optional) — Labels for computing the token classification loss. Indices should be in [0, ..., config.num_labels - 1].
Returns
loss (torch.FloatTensor of shape (1,), optional, returned when labels is provided) — Classification loss.
logits (torch.FloatTensor of shape (batch_size, sequence_length, config.num_labels)) — Classification scores (before SoftMax).
hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) — Tuple of torch.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 model at the output of each layer plus the optional initial embedding outputs.
attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) — Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).
Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
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 pre and post processing steps while the latter silently ignores them.
Example:
Copied
( config )
Parameters
MEGA Model with a span classification head on top for extractive question-answering tasks like SQuAD (a linear layers on top of the hidden-states output to compute span start logits and span end logits).
forward
Parameters
input_ids (torch.LongTensor of shape (batch_size, sequence_length)) — Indices of input sequence tokens in the vocabulary.
attention_mask (torch.FloatTensor of 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.
token_type_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) — Segment token indices to indicate first and second portions of the inputs. Indices are selected in [0,1]:
0 corresponds to a sentence A token,
1 corresponds to a sentence B token. This parameter can only be used when the model is initialized with add_token_type_embeddings parameter set to True. All the value in this tensor should be always < config.type_vocab_size.
inputs_embeds (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size), optional) — Optionally, instead of passing input_ids you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert input_ids indices into associated vectors than the model’s internal embedding lookup matrix.
output_attentions (bool, optional) — Whether or not to return the attentions tensors of all attention layers. See attentions under returned tensors for more detail.
output_hidden_states (bool, optional) — Whether or not to return the hidden states of all layers. See hidden_states under returned tensors for more detail.
start_positions (torch.LongTensor of shape (batch_size,), optional) — Labels for position (index) of the start of the labelled span for computing the token classification loss. Positions are clamped to the length of the sequence (sequence_length). Position outside of the sequence are not taken into account for computing the loss.
end_positions (torch.LongTensor of shape (batch_size,), optional) — Labels for position (index) of the end of the labelled span for computing the token classification loss. Positions are clamped to the length of the sequence (sequence_length). Position outside of the sequence are not taken into account for computing the loss.
Returns
loss (torch.FloatTensor of shape (1,), optional, returned when labels is provided) — Total span extraction loss is the sum of a Cross-Entropy for the start and end positions.
start_logits (torch.FloatTensor of shape (batch_size, sequence_length)) — Span-start scores (before SoftMax).
end_logits (torch.FloatTensor of shape (batch_size, sequence_length)) — Span-end scores (before SoftMax).
hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) — Tuple of torch.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 model at the output of each layer plus the optional initial embedding outputs.
attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) — Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).
Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
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 pre and post processing steps while the latter silently ignores them.
Example:
Copied
vocab_size (int, optional, defaults to 30522) — Vocabulary size of the Mega model. Defines the number of different tokens that can be represented by the inputs_ids passed when calling .
type_vocab_size (int, optional, defaults to 2) — The vocabulary size of the token_type_ids passed when calling . Only used if add_token_type_embeddings = True
This is the configuration class to store the configuration of a . It is used to instantiate a Mega model according to the specified arguments, defining the model architecture. Instantiating a configuration with the defaults will yield a similar configuration to that of the Mega architecture.
Configuration objects inherit from and can be used to control the model outputs. Read the documentation from for more information.
config () — 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 method to load the model weights.
This model inherits from . 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 subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.
.. _Mega: Moving Average Equipped Gated Attention:
( input_ids: typing.Optional[torch.Tensor] = Noneattention_mask: typing.Optional[torch.Tensor] = Nonetoken_type_ids: typing.Optional[torch.Tensor] = Noneinputs_embeds: typing.Optional[torch.Tensor] = Noneencoder_hidden_states: typing.Optional[torch.Tensor] = Noneencoder_attention_mask: typing.Optional[torch.Tensor] = Nonepast_key_values: typing.Optional[typing.List[torch.FloatTensor]] = Noneuse_cache: typing.Optional[bool] = Noneoutput_attentions: typing.Optional[bool] = Noneoutput_hidden_states: typing.Optional[bool] = Nonereturn_dict: typing.Optional[bool] = None ) → or tuple(torch.FloatTensor)
Indices can be obtained using . See and for details.
return_dict (bool, optional) — Whether or not to return a instead of a plain tuple.
or tuple(torch.FloatTensor)
A 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 () and inputs.
The forward method, overrides the __call__ special method.
config () — 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 method to load the model weights.
This model inherits from . 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 subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.
( input_ids: typing.Optional[torch.LongTensor] = Noneattention_mask: typing.Optional[torch.FloatTensor] = Nonetoken_type_ids: typing.Optional[torch.LongTensor] = Noneinputs_embeds: typing.Optional[torch.FloatTensor] = Noneencoder_hidden_states: typing.Optional[torch.FloatTensor] = Noneencoder_attention_mask: typing.Optional[torch.FloatTensor] = Nonelabels: typing.Optional[torch.LongTensor] = Nonepast_key_values: typing.Tuple[typing.Tuple[torch.FloatTensor]] = Noneuse_cache: typing.Optional[bool] = Noneoutput_attentions: typing.Optional[bool] = Noneoutput_hidden_states: typing.Optional[bool] = Nonereturn_dict: typing.Optional[bool] = None ) → or tuple(torch.FloatTensor)
Indices can be obtained using . See and for details.
return_dict (bool, optional) — Whether or not to return a instead of a plain tuple.
or tuple(torch.FloatTensor)
A 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 () and inputs.
The forward method, overrides the __call__ special method.
config () — 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 method to load the model weights.
This model inherits from . 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 subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.
( input_ids: typing.Optional[torch.LongTensor] = Noneattention_mask: typing.Optional[torch.FloatTensor] = Nonetoken_type_ids: typing.Optional[torch.LongTensor] = Noneinputs_embeds: typing.Optional[torch.FloatTensor] = Noneencoder_hidden_states: typing.Optional[torch.FloatTensor] = Noneencoder_attention_mask: typing.Optional[torch.FloatTensor] = Nonelabels: typing.Optional[torch.LongTensor] = Noneoutput_attentions: typing.Optional[bool] = Noneoutput_hidden_states: typing.Optional[bool] = Nonereturn_dict: typing.Optional[bool] = None ) → or tuple(torch.FloatTensor)
Indices can be obtained using . See and for details.
return_dict (bool, optional) — Whether or not to return a instead of a plain tuple.
or tuple(torch.FloatTensor)
A 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 () and inputs.
The forward method, overrides the __call__ special method.
config () — 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 method to load the model weights.
This model inherits from . 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 subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.
( input_ids: typing.Optional[torch.LongTensor] = Noneattention_mask: typing.Optional[torch.FloatTensor] = Nonetoken_type_ids: typing.Optional[torch.LongTensor] = Noneinputs_embeds: typing.Optional[torch.FloatTensor] = Nonelabels: typing.Optional[torch.LongTensor] = Noneoutput_attentions: typing.Optional[bool] = Noneoutput_hidden_states: typing.Optional[bool] = Nonereturn_dict: typing.Optional[bool] = None ) → or tuple(torch.FloatTensor)
Indices can be obtained using . See and for details.
return_dict (bool, optional) — Whether or not to return a instead of a plain tuple.
or tuple(torch.FloatTensor)
A 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 () and inputs.
The forward method, overrides the __call__ special method.
config () — 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 method to load the model weights.
This model inherits from . 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 subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.
( input_ids: typing.Optional[torch.LongTensor] = Nonetoken_type_ids: typing.Optional[torch.LongTensor] = Noneattention_mask: typing.Optional[torch.FloatTensor] = Nonelabels: typing.Optional[torch.LongTensor] = Noneinputs_embeds: typing.Optional[torch.FloatTensor] = Noneoutput_attentions: typing.Optional[bool] = Noneoutput_hidden_states: typing.Optional[bool] = Nonereturn_dict: typing.Optional[bool] = None ) → or tuple(torch.FloatTensor)
Indices can be obtained using . See and for details.
return_dict (bool, optional) — Whether or not to return a instead of a plain tuple.
or tuple(torch.FloatTensor)
A 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 () and inputs.
The forward method, overrides the __call__ special method.
config () — 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 method to load the model weights.
This model inherits from . 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 subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.
( input_ids: typing.Optional[torch.LongTensor] = Noneattention_mask: typing.Optional[torch.FloatTensor] = Nonetoken_type_ids: typing.Optional[torch.LongTensor] = Noneinputs_embeds: typing.Optional[torch.FloatTensor] = Nonelabels: typing.Optional[torch.LongTensor] = Noneoutput_attentions: typing.Optional[bool] = Noneoutput_hidden_states: typing.Optional[bool] = Nonereturn_dict: typing.Optional[bool] = None ) → or tuple(torch.FloatTensor)
Indices can be obtained using . See and for details.
return_dict (bool, optional) — Whether or not to return a instead of a plain tuple.
or tuple(torch.FloatTensor)
A 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 () and inputs.
The forward method, overrides the __call__ special method.
config () — 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 method to load the model weights.
This model inherits from . 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 subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.
( input_ids: typing.Optional[torch.LongTensor] = Noneattention_mask: typing.Optional[torch.FloatTensor] = Nonetoken_type_ids: typing.Optional[torch.LongTensor] = Noneinputs_embeds: typing.Optional[torch.FloatTensor] = Nonestart_positions: typing.Optional[torch.LongTensor] = Noneend_positions: typing.Optional[torch.LongTensor] = Noneoutput_attentions: typing.Optional[bool] = Noneoutput_hidden_states: typing.Optional[bool] = Nonereturn_dict: typing.Optional[bool] = None ) → or tuple(torch.FloatTensor)
Indices can be obtained using . See and for details.
return_dict (bool, optional) — Whether or not to return a instead of a plain tuple.
or tuple(torch.FloatTensor)
A 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 () and inputs.
The forward method, overrides the __call__ special method.