# RecurrentGemma ## Overview The Recurrent Gemma model was proposed in [RecurrentGemma: Moving Past Transformers for Efficient Open Language Models](https://huggingface.co/papers/2404.07839) 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: - The original checkpoints can be converted using the conversion script [`src/transformers/models/recurrent_gemma/convert_recurrent_gemma_weights_to_hf.py`](https://github.com/huggingface/transformers/blob/main/src/transformers/models/recurrent_gemma/convert_recurrent_gemma_to_hf.py). This model was contributed by [Arthur Zucker](https://huggingface.co/ArthurZ). The original code can be found [here](https://github.com/google-deepmind/recurrentgemma). ## RecurrentGemmaConfig[[transformers.RecurrentGemmaConfig]] #### transformers.RecurrentGemmaConfig[[transformers.RecurrentGemmaConfig]] [Source](https://github.com/huggingface/transformers/blob/v5.8.0/src/transformers/models/recurrent_gemma/configuration_recurrent_gemma.py#L25) This is the configuration class to store the configuration of a RecurrentGemmaModel. It is used to instantiate a Recurrent Gemma 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 [google/recurrentgemma-2b](https://huggingface.co/google/recurrentgemma-2b) Configuration objects inherit from [PreTrainedConfig](/docs/transformers/v5.8.0/en/main_classes/configuration#transformers.PreTrainedConfig) and can be used to control the model outputs. Read the documentation from [PreTrainedConfig](/docs/transformers/v5.8.0/en/main_classes/configuration#transformers.PreTrainedConfig) for more information. ```python >>> 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 ``` **Parameters:** num_hidden_layers (`int`, *optional*, defaults to `26`) : Number of hidden layers in the Transformer decoder. vocab_size (`int`, *optional*, defaults to `256000`) : Vocabulary size of the model. Defines the number of different tokens that can be represented by the `input_ids`. hidden_size (`int`, *optional*, defaults to `2560`) : Dimension of the hidden representations. intermediate_size (`int`, *optional*, defaults to `7680`) : Dimension of the MLP representations. num_attention_heads (`int`, *optional*, defaults to `10`) : Number of attention heads for each attention layer in the Transformer decoder. 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)`. attention_window_size (`int`, *optional*, defaults to 2048) : The size of the attention window used in the attention block. conv1d_width (`int`, *optional*, defaults to 4) : The kernel size of conv1d layers used in the recurrent blocks. logits_soft_cap (`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. rms_norm_eps (`float`, *optional*, defaults to `1e-06`) : The epsilon used by the rms normalization layers. 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` or when the model is a decoder-only generative model. pad_token_id (`int`, *optional*, defaults to `0`) : Token id used for padding in the vocabulary. eos_token_id (`Union[int, list[int]]`, *optional*, defaults to `1`) : Token id used for end-of-stream in the vocabulary. bos_token_id (`int`, *optional*, defaults to `2`) : Token id used for beginning-of-stream in the vocabulary. hidden_activation (`str`, *optional*, defaults to `gelu_pytorch_tanh`) : The non-linear activation function (function or string) in the decoder. For example, `"gelu"`, `"relu"`, `"silu"`, etc. rope_parameters (`Union[~modeling_rope_utils.RopeParameters, dict]`, *optional*) : Dictionary containing the configuration parameters for the RoPE embeddings. The dictionary should contain a value for `rope_theta` and optionally parameters used for scaling in case you want to use RoPE with longer `max_position_embeddings`. block_types (`list[str]`, *optional*, defaults to `('recurrent', 'recurrent', 'attention')`) : List of aleternating blocks that will be repeated to initialize the `temporal_block` layer. attention_dropout (`Union[float, int]`, *optional*, defaults to `0.0`) : The dropout ratio for the attention probabilities. num_key_value_heads (`int`, *optional*) : This is the number of key_value heads that should be used to implement Grouped Query Attention. If `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if `num_key_value_heads=1` the model will use Multi Query Attention (MQA) otherwise GQA is used. When converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed by meanpooling all the original heads within that group. For more details, check out [this paper](https://huggingface.co/papers/2305.13245). If it is not specified, will default to `num_attention_heads`. attention_bias (`bool`, *optional*, defaults to `False`) : Whether to use a bias in the query, key, value and output projection layers during self-attention. w_init_variance_scale (`float`, *optional*, defaults to 0.01) : weight initialization variance. tie_word_embeddings (`bool`, *optional*, defaults to `True`) : Whether to tie weight embeddings according to model's `tied_weights_keys` mapping. ## RecurrentGemmaModel[[transformers.RecurrentGemmaModel]] #### transformers.RecurrentGemmaModel[[transformers.RecurrentGemmaModel]] [Source](https://github.com/huggingface/transformers/blob/v5.8.0/src/transformers/models/recurrent_gemma/modeling_recurrent_gemma.py#L594) The bare Recurrent Gemma Model outputting raw hidden-states without any specific head on top. This model inherits from [PreTrainedModel](/docs/transformers/v5.8.0/en/main_classes/model#transformers.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](https://pytorch.org/docs/stable/nn.html#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. forwardtransformers.RecurrentGemmaModel.forwardhttps://github.com/huggingface/transformers/blob/v5.8.0/src/transformers/models/recurrent_gemma/modeling_recurrent_gemma.py#L613[{"name": "input_ids", "val": ": torch.LongTensor | None = None"}, {"name": "position_ids", "val": ": torch.LongTensor | None = None"}, {"name": "attention_mask", "val": ": torch.Tensor | None = None"}, {"name": "inputs_embeds", "val": ": torch.FloatTensor | None = None"}, {"name": "use_cache", "val": ": bool | None = None"}, {"name": "output_hidden_states", "val": ": bool | None = None"}, {"name": "return_dict", "val": ": bool | None = None"}, {"name": "past_key_values", "val": ": transformers.cache_utils.Cache | None = None"}, {"name": "**kwargs", "val": ": typing_extensions.Unpack[transformers.utils.generic.TransformersKwargs]"}]- **input_ids** (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*) -- Indices of input sequence tokens in the vocabulary. Padding will be ignored by default. Indices can be obtained using [AutoTokenizer](/docs/transformers/v5.8.0/en/model_doc/auto#transformers.AutoTokenizer). See [PreTrainedTokenizer.encode()](/docs/transformers/v5.8.0/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode) and [PreTrainedTokenizer.__call__()](/docs/transformers/v5.8.0/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__) for details. [What are input IDs?](../glossary#input-ids) - **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.n_positions - 1]`. [What are position IDs?](../glossary#position-ids) - **attention_mask** (`torch.Tensor` 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**. [What are attention masks?](../glossary#attention-mask) - **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. - **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`). - **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. - **return_dict** (`bool`, *optional*) -- Whether or not to return a [ModelOutput](/docs/transformers/v5.8.0/en/main_classes/output#transformers.utils.ModelOutput) instead of a plain tuple. - **past_key_values** (`~cache_utils.Cache`, *optional*) -- Pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention blocks) that can be used to speed up sequential decoding. This typically consists in the `past_key_values` returned by the model at a previous stage of decoding, when `use_cache=True` or `config.use_cache=True`. Only [Cache](/docs/transformers/v5.8.0/en/internal/generation_utils#transformers.Cache) instance is allowed as input, see our [kv cache guide](https://huggingface.co/docs/transformers/en/kv_cache). If no `past_key_values` are passed, [DynamicCache](/docs/transformers/v5.8.0/en/internal/generation_utils#transformers.DynamicCache) will be initialized by default. The model will output the same cache format that is fed as input. If `past_key_values` are used, the user is expected to input only unprocessed `input_ids` (those that don't have their past key value states given to this model) of shape `(batch_size, unprocessed_length)` instead of all `input_ids` of shape `(batch_size, sequence_length)`.0`BaseModelOutputWithNoAttention` or `tuple(torch.FloatTensor)`A `BaseModelOutputWithNoAttention` 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](/docs/transformers/v5.8.0/en/model_doc/recurrent_gemma#transformers.RecurrentGemmaConfig)) and inputs. The [RecurrentGemmaModel](/docs/transformers/v5.8.0/en/model_doc/recurrent_gemma#transformers.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. - **last_hidden_state** (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)`) -- Sequence of hidden-states at the output of the last layer of the model. - **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, num_channels, height, width)`. Hidden-states of the model at the output of each layer plus the optional initial embedding outputs. **Parameters:** config ([RecurrentGemmaConfig](/docs/transformers/v5.8.0/en/model_doc/recurrent_gemma#transformers.RecurrentGemmaConfig)) : 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()](/docs/transformers/v5.8.0/en/main_classes/model#transformers.PreTrainedModel.from_pretrained) method to load the model weights. **Returns:** ``BaseModelOutputWithNoAttention` or `tuple(torch.FloatTensor)`` A `BaseModelOutputWithNoAttention` 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](/docs/transformers/v5.8.0/en/model_doc/recurrent_gemma#transformers.RecurrentGemmaConfig)) and inputs. ## RecurrentGemmaForCausalLM[[transformers.RecurrentGemmaForCausalLM]] #### transformers.RecurrentGemmaForCausalLM[[transformers.RecurrentGemmaForCausalLM]] [Source](https://github.com/huggingface/transformers/blob/v5.8.0/src/transformers/models/recurrent_gemma/modeling_recurrent_gemma.py#L698) The Recurrent Gemma Model for causal language modeling. This model inherits from [PreTrainedModel](/docs/transformers/v5.8.0/en/main_classes/model#transformers.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](https://pytorch.org/docs/stable/nn.html#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. forwardtransformers.RecurrentGemmaForCausalLM.forwardhttps://github.com/huggingface/transformers/blob/v5.8.0/src/transformers/models/recurrent_gemma/modeling_recurrent_gemma.py#L710[{"name": "input_ids", "val": ": torch.LongTensor | None = None"}, {"name": "position_ids", "val": ": torch.LongTensor | None = None"}, {"name": "attention_mask", "val": ": torch.Tensor | None = None"}, {"name": "inputs_embeds", "val": ": torch.FloatTensor | None = None"}, {"name": "labels", "val": ": torch.LongTensor | None = None"}, {"name": "output_hidden_states", "val": ": bool | None = None"}, {"name": "return_dict", "val": ": bool | None = None"}, {"name": "use_cache", "val": ": bool | None = None"}, {"name": "logits_to_keep", "val": ": int | torch.Tensor = 0"}, {"name": "past_key_values", "val": ": transformers.cache_utils.Cache | None = None"}, {"name": "**kwargs", "val": ": typing_extensions.Unpack[transformers.utils.generic.TransformersKwargs]"}]- **input_ids** (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*) -- Indices of input sequence tokens in the vocabulary. Padding will be ignored by default. Indices can be obtained using [AutoTokenizer](/docs/transformers/v5.8.0/en/model_doc/auto#transformers.AutoTokenizer). See [PreTrainedTokenizer.encode()](/docs/transformers/v5.8.0/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode) and [PreTrainedTokenizer.__call__()](/docs/transformers/v5.8.0/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__) for details. [What are input IDs?](../glossary#input-ids) - **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.n_positions - 1]`. [What are position IDs?](../glossary#position-ids) - **attention_mask** (`torch.Tensor` 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**. [What are attention masks?](../glossary#attention-mask) - **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. - **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]`. - **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. - **return_dict** (`bool`, *optional*) -- Whether or not to return a [ModelOutput](/docs/transformers/v5.8.0/en/main_classes/output#transformers.utils.ModelOutput) instead of a plain tuple. - **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`). - **logits_to_keep** (`Union[int, torch.Tensor]`, *optional*, defaults to `0`) -- If an `int`, compute logits for the last `logits_to_keep` tokens. If `0`, calculate logits for all `input_ids` (special case). Only last token logits are needed for generation, and calculating them only for that token can save memory, which becomes pretty significant for long sequences or large vocabulary size. If a `torch.Tensor`, must be 1D corresponding to the indices to keep in the sequence length dimension. This is useful when using packed tensor format (single dimension for batch and sequence length). - **past_key_values** (`~cache_utils.Cache`, *optional*) -- Pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention blocks) that can be used to speed up sequential decoding. This typically consists in the `past_key_values` returned by the model at a previous stage of decoding, when `use_cache=True` or `config.use_cache=True`. Only [Cache](/docs/transformers/v5.8.0/en/internal/generation_utils#transformers.Cache) instance is allowed as input, see our [kv cache guide](https://huggingface.co/docs/transformers/en/kv_cache). If no `past_key_values` are passed, [DynamicCache](/docs/transformers/v5.8.0/en/internal/generation_utils#transformers.DynamicCache) will be initialized by default. The model will output the same cache format that is fed as input. If `past_key_values` are used, the user is expected to input only unprocessed `input_ids` (those that don't have their past key value states given to this model) of shape `(batch_size, unprocessed_length)` instead of all `input_ids` of shape `(batch_size, sequence_length)`.0[CausalLMOutput](/docs/transformers/v5.8.0/en/main_classes/output#transformers.modeling_outputs.CausalLMOutput) or `tuple(torch.FloatTensor)`A [CausalLMOutput](/docs/transformers/v5.8.0/en/main_classes/output#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](/docs/transformers/v5.8.0/en/model_doc/recurrent_gemma#transformers.RecurrentGemmaConfig)) and inputs. The [RecurrentGemmaForCausalLM](/docs/transformers/v5.8.0/en/model_doc/recurrent_gemma#transformers.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. - **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. Example: ```python >>> 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?" ``` **Parameters:** config ([RecurrentGemmaForCausalLM](/docs/transformers/v5.8.0/en/model_doc/recurrent_gemma#transformers.RecurrentGemmaForCausalLM)) : 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()](/docs/transformers/v5.8.0/en/main_classes/model#transformers.PreTrainedModel.from_pretrained) method to load the model weights. **Returns:** `[CausalLMOutput](/docs/transformers/v5.8.0/en/main_classes/output#transformers.modeling_outputs.CausalLMOutput) or `tuple(torch.FloatTensor)`` A [CausalLMOutput](/docs/transformers/v5.8.0/en/main_classes/output#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](/docs/transformers/v5.8.0/en/model_doc/recurrent_gemma#transformers.RecurrentGemmaConfig)) and inputs.