The Recurrent Gemma model was proposed in RecurrentGemma: Moving Past Transformers for Efficient Open Language Models by the Griffin, RLHF and Gemma Teams of Google.
The abstract from the paper is the following:
We introduce RecurrentGemma, an open language model which uses Google’s novel Griffin architecture. Griffin combines linear recurrences with local attention to achieve excellent performance on language. It has a fixed-sized state, which reduces memory use and enables efficient inference on long sequences. We provide a pre-trained model with 2B non-embedding parameters, and an instruction tuned variant. Both models achieve comparable performance to Gemma-2B despite being trained on fewer tokens.
Tips:
src/transformers/models/recurrent_gemma/convert_recurrent_gemma_weights_to_hf.py
.This model was contributed by Arthur Zucker. The original code can be found here.
( num_hidden_layers = 26 vocab_size = 256000 hidden_size = 2560 intermediate_size = 7680 num_attention_heads = 10 lru_width = None attention_window_size = 2048 conv1d_width = 4 logits_soft_cap = 30.0 rms_norm_eps = 1e-06 use_cache = True pad_token_id = 0 eos_token_id = 1 bos_token_id = 2 hidden_activation = 'gelu_pytorch_tanh' partial_rotary_factor = 0.5 rope_theta = 10000.0 block_types = ('recurrent', 'recurrent', 'attention') attention_dropout = 0.0 num_key_value_heads = None attention_bias = False w_init_variance_scale = 0.01 **kwargs )
Parameters
int
, optional, defaults to 26) —
The number of hidden layers in the model. int
, optional, defaults to 256000) —
Vocabulary size of the RecurrentGemma model. Defines the number of
different tokens that can be represented by the
inputs_ids
passed when calling RecurrentGemmaModel int
, optional, defaults to 2560) —
Dimension of the hidden representations. int
, optional, defaults to 7680) —
Dimension of the MLP representations. int
, optional, defaults to 10) —
The number of heads for the attention block and the number of
heads/blocks for the block-diagonal layers used in the RG-LRU gates.
This number must divide hidden_size
and lru_width
. int
or None
, optional) —
Dimension of the hidden representations of the RG-LRU. If None
this will be set to hidden_size
.
Whether to scale the output of the embeddings by sqrt(hidden_size)
. int
, optional, defaults to 2048) —
The size of the attention window used in the attention block. int
, optional, defaults to 4) —
The kernel size of conv1d layers used in the recurrent blocks. float
, optional, defaults to 30.0) —
The value at which the logits should be soft-capped to after the transformer and LM-head computation in the Causal LM architecture. float
, optional, defaults to 1e-06) —
The epsilon used by the rms normalization layers. bool
, optional, defaults to True
) —
Whether the model should return the last key/values
attentions (not used by all models). Only
relevant if config.is_decoder=True
. int
, optional, defaults to 0) —
Padding token id. int
, optional, defaults to 1) —
End of stream token id. int
, optional, defaults to 2) —
Beginning of stream token id. str` or `function
, optional, defaults to "gelu_pytorch_tanh"
) —
The hidden activation used in the recurrent block as well as the MLP layer of the decoder layers. float
, optional, defaults to 0.5) —
The partial rotary factor used in the initialization of the rotary embeddings. float
, optional, defaults to 10000.0) —
The base period of the RoPE embeddings. List[str]
, optional, defaults to ('recurrent', 'recurrent', 'attention')
) —
List of aleternating blocks that will be repeated to initialize the temporal_block
layer. float
, optional, defaults to 0.0) — dropout value to use after the attention softmax. 16
, optional, defaults to 16) — Number of key value heads to use GQA. bool
, optional, defaults to False
) — whether or not the linear q,k,v of the Attention layer should have bias float
, optional, defaults to 0.01) — weight initialization variance. This is the configuration class to store the configuration of a RecurrentGemmaModel. It is used to instantiate a RecurrentGemma 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 RecurrentGemma-7B.
Configuration objects inherit from PretrainedConfig and can be used to control the model outputs. Read the documentation from PretrainedConfig for more information.
>>> from transformers import RecurrentGemmaModel, RecurrentGemmaConfig
>>> # Initializing a RecurrentGemma recurrentgemma-2b style configuration
>>> configuration = RecurrentGemmaConfig()
>>> # Initializing a model from the recurrentgemma-2b style configuration
>>> model = RecurrentGemmaModel(configuration)
>>> # Accessing the model configuration
>>> configuration = model.config
( config: RecurrentGemmaConfig )
Parameters
The bare RecurrentGemma 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.
Transformer decoder consisting of config.num_hidden_layers layers. Each layer is a RecurrentGemmaDecoderLayer
( input_ids: LongTensor = None position_ids: Optional = None attention_mask: Optional = None cache_position: Optional = None inputs_embeds: Optional = None use_cache: Optional = None output_hidden_states: Optional = None return_dict: Optional = None **kwargs )
Parameters
torch.LongTensor
of shape (batch_size, sequence_length)
) —
Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide
it.
Indices can be obtained using AutoTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.
torch.Tensor
of shape (batch_size, sequence_length)
, optional) —
Mask to avoid performing attention on padding token indices. Mask values selected in [0, 1]
:
Indices can be obtained using AutoTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.
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.n_positions - 1]
.
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. 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
). bool
, optional) —
Whether or not to return the hidden states of all See hidden_states
under returned tensors for
more detail. bool
, optional) —
Whether or not to return a ModelOutput instead of a plain tuple. torch.LongTensor
of shape (sequence_length)
, optional) —
Indices depicting the position of the input sequence tokens in the sequence. Contrarily to position_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. The RecurrentGemmaModel 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.
( input_ids: Optional = None cache_position: Optional = None attention_mask: Optional = None inputs_embeds: Optional = None labels: Optional = None output_hidden_states: Optional = None return_dict: Optional = None use_cache: Optional = None **kwargs ) → transformers.modeling_outputs.CausalLMOutput or tuple(torch.FloatTensor)
Parameters
torch.LongTensor
of shape (batch_size, sequence_length)
) —
Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide
it.
Indices can be obtained using AutoTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.
torch.Tensor
of shape (batch_size, sequence_length)
, optional) —
Mask to avoid performing attention on padding token indices. Mask values selected in [0, 1]
:
Indices can be obtained using AutoTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.
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.n_positions - 1]
.
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. 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
). bool
, optional) —
Whether or not to return the hidden states of all See hidden_states
under returned tensors for
more detail. bool
, optional) —
Whether or not to return a ModelOutput instead of a plain tuple. torch.LongTensor
of shape (sequence_length)
, optional) —
Indices depicting the position of the input sequence tokens in the sequence. Contrarily to position_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.
Args —
labels (torch.LongTensor
of shape (batch_size, sequence_length)
, optional):
Labels for computing the masked language modeling loss. Indices should either be in [0, ..., config.vocab_size]
or -100 (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]
.
Returns
transformers.modeling_outputs.CausalLMOutput or tuple(torch.FloatTensor)
A transformers.modeling_outputs.CausalLMOutput 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 (RecurrentGemmaConfig) and inputs.
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.
The RecurrentGemmaForCausalLM 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, RecurrentGemmaForCausalLM
>>> model = RecurrentGemmaForCausalLM.from_pretrained("google/recurrentgemma-2b")
>>> tokenizer = AutoTokenizer.from_pretrained("google/recurrentgemma-2b")
>>> prompt = "What is your favorite condiment?"
>>> inputs = tokenizer(prompt, return_tensors="pt")
>>> # Generate
>>> generate_ids = model.generate(inputs.input_ids, max_length=30)
>>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
"What is your favorite condiment?"