GPT-NeoX-Japanese
Overview
We introduce GPT-NeoX-Japanese, which is an autoregressive language model for Japanese, trained on top of https://github.com/EleutherAI/gpt-neox. Japanese is a unique language with its large vocabulary and a combination of hiragana, katakana, and kanji writing scripts. To address this distinct structure of the Japanese language, we use a special sub-word tokenizer. We are very grateful to tanreinama for open-sourcing this incredibly helpful tokenizer. Following the recommendations from Google’s research on PaLM, we have removed bias parameters from transformer blocks, achieving better model performance. Please refer this article in detail.
Development of the model was led by Shinya Otani, Takayoshi Makabe, Anuj Arora, and Kyo Hattori from ABEJA, Inc.. For more information on this model-building activity, please refer here (ja).
Usage example
The generate()
method can be used to generate text using GPT NeoX Japanese model.
>>> from transformers import GPTNeoXJapaneseForCausalLM, GPTNeoXJapaneseTokenizer
>>> model = GPTNeoXJapaneseForCausalLM.from_pretrained("abeja/gpt-neox-japanese-2.7b")
>>> tokenizer = GPTNeoXJapaneseTokenizer.from_pretrained("abeja/gpt-neox-japanese-2.7b")
>>> prompt = "人とAIが協調するためには、"
>>> input_ids = tokenizer(prompt, return_tensors="pt").input_ids
>>> gen_tokens = model.generate(
... input_ids,
... do_sample=True,
... temperature=0.9,
... max_length=100,
... )
>>> gen_text = tokenizer.batch_decode(gen_tokens, skip_special_tokens=True)[0]
>>> print(gen_text)
人とAIが協調するためには、AIと人が共存し、AIを正しく理解する必要があります。
Resources
GPTNeoXJapaneseConfig
class transformers.GPTNeoXJapaneseConfig
< source >( vocab_size = 32000 hidden_size = 2560 num_hidden_layers = 32 num_attention_heads = 32 intermediate_multiple_size = 4 hidden_act = 'gelu' rotary_pct = 1.0 rotary_emb_base = 10000 max_position_embeddings = 2048 initializer_range = 0.02 layer_norm_eps = 1e-05 use_cache = True bos_token_id = 31996 eos_token_id = 31999 attention_dropout = 0.1 hidden_dropout = 0.0 **kwargs )
Parameters
- vocab_size (
int
, optional, defaults to 32000) — Vocabulary size of the GPTNeoXJapanese model. Defines the number of different tokens that can be represented by theinputs_ids
passed when callingGPTNeoXJapanese
. - hidden_size (
int
, optional, defaults to 2560) — Dimension of the encoder layers and the pooler layer. - num_hidden_layers (
int
, optional, defaults to 32) — Number of hidden layers in the Transformer encoder. - num_attention_heads (
int
, optional, defaults to 32) — Number of attention heads for each attention layer in the Transformer encoder. - intermediate_multiple_size (
int
, optional, defaults to 4) — Dimension of the “intermediate” layer in the Transformer encoder is calculated by hidden_size * intermediate_multiple_size. - hidden_act (
str
orfunction
, optional, defaults to"gelu"
) — The non-linear activation function (function or string) in the encoder and pooler. - rotary_pct (
float
, optional, defaults to 1.00) — percentage of hidden dimensions to allocate to rotary embeddings - rotary_emb_base (
int
, optional, defaults to 10000) — base for computing rotary embeddings frequency - max_position_embeddings (
int
, optional, defaults to 2048) — The maximum sequence length that this model might ever be used with. - initializer_range (
float
, optional, defaults to 0.02) — The standard deviation of the truncated_normal_initializer for initializing all weight matrices. - layer_norm_eps (
float
, optional, defaults to 1e-5) — The epsilon used by the layer normalization layers. - use_cache (
bool
, optional, defaults toTrue
) — Whether or not the model should return the last key/values attentions (not used by all models). Only relevant ifconfig.is_decoder=True
. - attention_dropout (
float
, optional, defaults to 0.1) — The dropout ratio for the attention. - hidden_dropout (
float
, optional, defaults to 0.0) — The dropout ratio for the hidden layer. Example —
This is the configuration class to store the configuration of a GPTNeoXModelJapanese
. It is used to instantiate
a GPTNeoX 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 GPTNeoXJapanese
abeja/gpt-neox-japanese-2.7b architecture.
Configuration objects inherit from PretrainedConfig and can be used to control the model outputs. Read the documentation from PretrainedConfig for more information. Default configs is set as 2.7B model
>>> from transformers import GPTNeoXJapaneseConfig, GPTNeoXJapaneseModel
>>> # Initializing a GPTNeoXJapanese gpt-neox-japanese-2.7b style configuration
>>> configuration = GPTNeoXJapaneseConfig()
>>> # Initializing a model (with random weights) from the gpt-neox-japanese-2.7b style configuration
>>> model = GPTNeoXJapaneseModel(configuration)
>>> # Accessing the model configuration
>>> configuration = model.config
GPTNeoXJapaneseTokenizer
class transformers.GPTNeoXJapaneseTokenizer
< source >( vocab_file emoji_file unk_token = '<|endoftext|>' pad_token = '<|endoftext|>' bos_token = '<|startoftext|>' eos_token = '<|endoftext|>' do_clean_text = False **kwargs )
Parameters
- vocab_file (
str
) — File containing the vocabulary. - emoji_file (
str
) — File containing the emoji. - unk_token (
str
, optional, defaults to"<|endoftext|>"
) — The unknown token. A token that is not in the vocabulary cannot be converted to an ID and is set to be this token instead. - pad_token (
str
, optional, defaults to"<|endoftext|>"
) — The token used for padding - bos_token (
str
, optional, defaults to"<|startoftext|>"
) — The beginning of sequence token. - eos_token (
str
, optional, defaults to"<|endoftext|>"
) — The end of sequence token. - do_clean_text (
bool
, optional, defaults toFalse
) — Whether or not to clean text for URL, EMAIL, TEL, Japanese DATE and Japanese PRICE.
This tokenizer inherits from PreTrainedTokenizer and is based on Japanese special Sub-Word-Encoding that is used in this repository (https://github.com/tanreinama/Japanese-BPEEncoder_V2). Check the repository for details. Japanese has a relatively large vocabulary and there is no separation between words. Furthermore, the language is a combination of hiragana, katakana, and kanji, and variants such as “1” and “①” are often used. In order to cope with these, this tokenizer has the following features
- Subword-by-subword segmentation, which is intermediate between byte strings and morphological analysis.
- BPEs are created for each Kanji, Hiragana, and Katakana character, and there are no BPEs that cross character types, such as Kanji + Hiragana or Hiragana + Katakana.
- All-byte encoding that does not require <unk>.
- Independent of UTF codes such as 2-byte and 3-byte characters
- Conversion of heterographs to the same token_id
- Emoji and Emoticon are grouped into 12 types as special tags.
Example:
>>> from transformers import GPTNeoXJapaneseTokenizer
>>> tokenizer = GPTNeoXJapaneseTokenizer.from_pretrained("abeja/gpt-neox-japanese-2.7b")
>>> # You can confirm both 慶応 and 慶應 are encoded to 17749
>>> tokenizer("吾輩は猫である🐯。実は慶応(慶應)大学出身")["input_ids"]
[30014, 26883, 26638, 27228, 25, 26650, 31732, 31679, 27809, 26638, 17749, 31592, 17749, 31593, 321, 1281]
>>> # Both 慶応 and 慶應 are decoded to 慶応
>>> tokenizer.decode(tokenizer("吾輩は猫である🐯。実は慶応(慶應)大学出身")["input_ids"])
'吾輩は猫である🐯。実は慶応(慶応)大学出身'
Converts a sequence of tokens (string) in a single string.
GPTNeoXJapaneseModel
class transformers.GPTNeoXJapaneseModel
< source >( config )
Parameters
- config (~GPTNeoXJapaneseConfig) — 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.
The bare GPTNeoXJapanese Model transformer outputting raw hidden-states without any specific head on top. This model is a PyTorch torch.nn.Module sub-class. 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: Optional = None attention_mask: Optional = None head_mask: Optional = None inputs_embeds: Optional = None past_key_values: Optional = None use_cache: Optional = None output_attentions: Optional = None output_hidden_states: Optional = None return_dict: Optional = None ) → transformers.modeling_outputs.BaseModelOutputWithPast or tuple(torch.FloatTensor)
Parameters
- input_ids (
torch.LongTensor
of shape(batch_size, sequence_length)
) — Indices of input sequence tokens in the vocabulary.Indices can be obtained using AutoTokenizer.
- 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.
- position_ids (
torch.LongTensor
of shape(batch_size, sequence_length)
, optional) — Indices of positions of each input sequence tokens in the position embeddings. Selected in the range[0, config.max_position_embeddings - 1]
. - head_mask (
torch.FloatTensor
of shape(num_heads,)
or(num_layers, num_heads)
, optional) — Mask to nullify selected heads of the self-attention modules. Mask values selected in[0, 1]
:- 1 indicates the head is not masked,
- 0 indicates the head is masked.
- inputs_embeds (
torch.FloatTensor
of shape(batch_size, sequence_length, hidden_size)
, optional) — Optionally, instead of passinginput_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. Seeattentions
under returned tensors for more detail. - output_hidden_states (
bool
, optional) — Whether or not to return the hidden states of all layers. Seehidden_states
under returned tensors for more detail. - return_dict (
bool
, optional) — Whether or not to return a ModelOutput instead of a plain tuple. - past_key_values (
tuple(tuple(torch.FloatTensor))
of lengthconfig.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. Ifpast_key_values
are used, the user can optionally input only the lastdecoder_input_ids
(those that don’t have their past key value states given to this model) of shape(batch_size, 1)
instead of alldecoder_input_ids
of shape(batch_size, sequence_length)
. - use_cache (
bool
, optional) — If set toTrue
,past_key_values
key value states are returned and can be used to speed up decoding (seepast_key_values
).
Returns
transformers.modeling_outputs.BaseModelOutputWithPast or tuple(torch.FloatTensor)
A transformers.modeling_outputs.BaseModelOutputWithPast 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 (GPTNeoXJapaneseConfig) and inputs.
-
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.If
past_key_values
is used only the last hidden-state of the sequences of shape(batch_size, 1, hidden_size)
is output. -
past_key_values (
tuple(tuple(torch.FloatTensor))
, optional, returned whenuse_cache=True
is passed or whenconfig.use_cache=True
) — Tuple oftuple(torch.FloatTensor)
of lengthconfig.n_layers
, with each tuple having 2 tensors of shape(batch_size, num_heads, sequence_length, embed_size_per_head)
) and optionally ifconfig.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 (seepast_key_values
input) to speed up sequential decoding. -
hidden_states (
tuple(torch.FloatTensor)
, optional, returned whenoutput_hidden_states=True
is 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 model at the output of each layer plus the optional initial embedding outputs.
-
attentions (
tuple(torch.FloatTensor)
, optional, returned whenoutput_attentions=True
is 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 after the attention softmax, used to compute the weighted average in the self-attention heads.
The GPTNeoXJapaneseModel forward method, overrides the __call__
special method.
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:
>>> from transformers import AutoTokenizer, GPTNeoXJapaneseModel
>>> import torch
>>> tokenizer = AutoTokenizer.from_pretrained("abeja/gpt-neox-japanese-2.7b")
>>> model = GPTNeoXJapaneseModel.from_pretrained("abeja/gpt-neox-japanese-2.7b")
>>> inputs = tokenizer("日本語のGPT-neoxがHugging Faceで使えます😀", return_tensors="pt")
>>> outputs = model(**inputs)
>>> last_hidden_states = outputs.last_hidden_state
GPTNeoXJapaneseForCausalLM
class transformers.GPTNeoXJapaneseForCausalLM
< source >( config )
Parameters
- config (~GPTNeoXJapaneseConfig) — 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.
GPTNeoXJapanese Model with a language modeling
head on top for Classifier Model fine-tuning.
This model is a PyTorch torch.nn.Module sub-class. 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: Optional = None attention_mask: Optional = None inputs_embeds: Optional = None head_mask: Optional = None past_key_values: Optional = None labels: Optional = None use_cache: Optional = None output_attentions: Optional = None output_hidden_states: Optional = None return_dict: Optional = None ) → transformers.modeling_outputs.CausalLMOutputWithPast or tuple(torch.FloatTensor)
Parameters
- input_ids (
torch.LongTensor
of shape(batch_size, sequence_length)
) — Indices of input sequence tokens in the vocabulary.Indices can be obtained using AutoTokenizer.
- 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.
- position_ids (
torch.LongTensor
of shape(batch_size, sequence_length)
, optional) — Indices of positions of each input sequence tokens in the position embeddings. Selected in the range[0, config.max_position_embeddings - 1]
. - head_mask (
torch.FloatTensor
of shape(num_heads,)
or(num_layers, num_heads)
, optional) — Mask to nullify selected heads of the self-attention modules. Mask values selected in[0, 1]
:- 1 indicates the head is not masked,
- 0 indicates the head is masked.
- inputs_embeds (
torch.FloatTensor
of shape(batch_size, sequence_length, hidden_size)
, optional) — Optionally, instead of passinginput_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. Seeattentions
under returned tensors for more detail. - output_hidden_states (
bool
, optional) — Whether or not to return the hidden states of all layers. Seehidden_states
under returned tensors for more detail. - return_dict (
bool
, optional) — Whether or not to return a ModelOutput instead of a plain tuple. - past_key_values (
tuple(tuple(torch.FloatTensor))
, optional, returned whenuse_cache=True
is passed or whenconfig.use_cache=True
) — Tuple oftuple(torch.FloatTensor)
of lengthconfig.n_layers
, with each tuple having 2 tensors of shape(batch_size, num_heads, sequence_length, embed_size_per_head)
) and 2 additional tensors of shape(batch_size, num_heads, encoder_sequence_length, embed_size_per_head)
. The two additional tensors are only required when the model is used as a decoder in a Sequence to Sequence model.Contains pre-computed hidden-states (key and values in the self-attention blocks that can be used (see
past_key_values
input) to speed up sequential decoding.If
past_key_values
are used, the user can optionally input only the lastdecoder_input_ids
(those that don’t have their past key value states given to this model) of shape(batch_size, 1)
instead of alldecoder_input_ids
of shape(batch_size, sequence_length)
. - 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]
(seeinput_ids
docstring) Tokens with indices set to-100
are ignored (masked), the loss is only computed for the tokens with labels n[0, ..., config.vocab_size]
. - use_cache (
bool
, optional) — If set toTrue
,past_key_values
key value states are returned and can be used to speed up decoding (seepast_key_values
).
Returns
transformers.modeling_outputs.CausalLMOutputWithPast or tuple(torch.FloatTensor)
A transformers.modeling_outputs.CausalLMOutputWithPast 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 (GPTNeoXJapaneseConfig) and inputs.
-
loss (
torch.FloatTensor
of shape(1,)
, optional, returned whenlabels
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). -
past_key_values (
tuple(tuple(torch.FloatTensor))
, optional, returned whenuse_cache=True
is passed or whenconfig.use_cache=True
) — Tuple oftuple(torch.FloatTensor)
of lengthconfig.n_layers
, with each tuple having 2 tensors of shape(batch_size, num_heads, sequence_length, embed_size_per_head)
)Contains pre-computed hidden-states (key and values in the self-attention blocks) that can be used (see
past_key_values
input) to speed up sequential decoding. -
hidden_states (
tuple(torch.FloatTensor)
, optional, returned whenoutput_hidden_states=True
is 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 model at the output of each layer plus the optional initial embedding outputs.
-
attentions (
tuple(torch.FloatTensor)
, optional, returned whenoutput_attentions=True
is 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 after the attention softmax, used to compute the weighted average in the self-attention heads.
The GPTNeoXJapaneseForCausalLM forward method, overrides the __call__
special method.
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:
>>> from transformers import AutoTokenizer, GPTNeoXJapaneseForCausalLM, GPTNeoXJapaneseConfig
>>> import torch
>>> tokenizer = AutoTokenizer.from_pretrained("abeja/gpt-neox-japanese-2.7b")
>>> config = GPTNeoXJapaneseConfig.from_pretrained("abeja/gpt-neox-japanese-2.7b")
>>> config.is_decoder = True
>>> model = GPTNeoXJapaneseForCausalLM.from_pretrained("abeja/gpt-neox-japanese-2.7b", config=config)
>>> inputs = tokenizer("日本語のGPT-neoxがHugging Faceで使えます😀", return_tensors="pt")
>>> outputs = model(**inputs)
>>> prediction_logits = outputs.logits