Zamba
Zamba is a large language model (LLM) trained by Zyphra, and made available under an Apache 2.0 license. Please see the Zyphra Hugging Face repository for model weights.
This model was contributed by pglo.
Model details
Zamba-7B-v1 is a hybrid between state-space models (Specifically Mamba) and transformer, and was trained using next-token prediction. Zamba uses a shared transformer layer after every 6 mamba blocks. It uses the Mistral v0.1 tokenizer. We came to this architecture after a series of ablations at small scales. Zamba-7B-v1 was pre-trained on 1T tokens of text and code data.
Quick start
Presequities
Zamba requires you use transformers
version 4.46.0 or higher:
pip install transformers>=4.45.0
In order to run optimized Mamba implementations, you first need to install mamba-ssm
and causal-conv1d
:
pip install mamba-ssm causal-conv1d>=1.2.0
You also have to have the model on a CUDA device.
You can run the model not using the optimized Mamba kernels, but it is not recommended as it will result in significantly lower latencies. In order to do that, you’ll need to specify use_mamba_kernels=False
when loading the model.
Inference
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
tokenizer = AutoTokenizer.from_pretrained("Zyphra/Zamba-7B-v1")
model = AutoModelForCausalLM.from_pretrained("Zyphra/Zamba-7B-v1", device_map="auto", torch_dtype=torch.bfloat16)
input_text = "A funny prompt would be "
input_ids = tokenizer(input_text, return_tensors="pt").to("cuda")
outputs = model.generate(**input_ids, max_new_tokens=100)
print(tokenizer.decode(outputs[0]))
Model card
The model cards can be found at:
Issues
For issues with model output, or community discussion, please use the Hugging Face community forum
License
The model weights are open-sourced via an Apache 2.0 license.
ZambaConfig
class transformers.ZambaConfig
< source >( vocab_size = 32000 tie_word_embeddings = True hidden_size = 3712 attention_hidden_size = None intermediate_size = 14848 num_hidden_layers = 76 num_attention_heads = 16 attention_head_dim = None num_key_value_heads = 16 n_mamba_heads = 2 hidden_act = 'gelu' hidden_mamba_act = 'silu' initializer_range = 0.02 rms_norm_eps = 1e-05 use_cache = True num_logits_to_keep = 1 pad_token_id = 0 bos_token_id = 1 eos_token_id = 2 max_position_embeddings = 4096 attention_dropout = 0.0 attn_layer_period = 6 attn_layer_offset = 4 use_mamba_kernels = True mamba_d_state = 16 mamba_d_conv = 4 mamba_expand = 2 mamba_dt_rank = 'auto' time_step_min = 0.001 time_step_max = 0.1 time_step_floor = 0.0001 mamba_conv_bias = True mamba_proj_bias = False **kwargs )
Parameters
- vocab_size (
int
, optional, defaults to 32000) — Vocabulary size of the Zamba model. Defines the number of different tokens that can be represented by theinputs_ids
passed when calling ZambaModel - tie_word_embeddings (
bool
, optional, defaults toTrue
) — Whether the model’s input and output word embeddings should be tied. Note that this is only relevant if the model has a output word embedding layer. - hidden_size (
int
, optional, defaults to 3712) — Dimension of the hidden representations. - attention_hidden_size (
int
, optional) — Dimension of the hidden representations of the inputs to the Attention layer. - intermediate_size (
int
, optional, defaults to 14848) — Dimension of the MLP representations. - num_hidden_layers (
int
, optional, defaults to 76) — Number of hidden layers in the model. - num_attention_heads (
int
, optional, defaults to 16) — Number of attention heads for each attention layer in the Transformer decoder. - attention_head_dim (
int
, optional) — Dimension of the attention head in the Transformer decoder. - num_key_value_heads (
int
, optional, defaults to 16) — This is the number of key_value heads that should be used to implement Grouped Query Attention. Ifnum_key_value_heads=None
, 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 checkout this paper. - n_mamba_heads (
int
, optional, defaults to 2) — Number of mamba heads for each mamba layer. - hidden_act (
str
orfunction
, optional, defaults to"gelu"
) — The non-linear activation function (function or string) in the decoder. - hidden_mamba_act (
str
orfunction
, optional, defaults to"silu"
) — The non-linear activation function (function or string) in the mamba layer. - initializer_range (
float
, optional, defaults to 0.02) — The standard deviation of the truncated_normal_initializer for initializing all weight matrices. - rms_norm_eps (
float
, optional, defaults to 1e-05) — The epsilon used by the rms 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
. - num_logits_to_keep (
int
orNone
, optional, defaults to 1) — Number of prompt logits to calculate during generation. IfNone
, all logits will be calculated. If an integer value, only lastnum_logits_to_keep
logits will be calculated. Default is 1 because only the logits of the last prompt token are needed for generation. For long sequences, the logits for the entire sequence may use a lot of memory so, settingnum_logits_to_keep=1
will reduce memory footprint significantly. - pad_token_id (
int
, optional, defaults to 0) — The id of the padding token. - bos_token_id (
int
, optional, defaults to 1) — The id of the “beginning-of-sequence” token. - eos_token_id (
int
, optional, defaults to 2) — The id of the “end-of-sequence” token. - max_position_embeddings (
int
, optional, defaults to 4096) — This value doesn’t have any real effect. The maximum sequence length that this model is intended to be used with. It can be used with longer sequences, but performance may degrade. - attention_dropout (
float
, optional, defaults to 0.0) — The dropout ratio for the attention probabilities. - attn_layer_period (
int
, optional, defaults to 6) — Once in this many layers, we will have a shared attention layer - attn_layer_offset (
int
, optional, defaults to 4) — Offset of the shared attention layer - use_mamba_kernels (
bool
, optional, defaults toTrue
) — Flag indicating whether or not to use the fast mamba kernels. These are available only ifmamba-ssm
andcausal-conv1d
are installed, and the mamba modules are running on a CUDA device. Raises ValueError ifTrue
and kernels are not available - mamba_d_state (
int
, optional, defaults to 16) — The dimension the mamba state space latents - mamba_d_conv (
int
, optional, defaults to 4) — The size of the mamba convolution kernel - mamba_expand (
int
, optional, defaults to 2) — Expanding factor (relative to hidden_size) used to determine the mamba intermediate size - mamba_dt_rank (
Union[int,str]
, optional, defaults to"auto"
) — Rank of the the mamba discretization projection matrix."auto"
means that it will default tomath.ceil(self.hidden_size / 16)
- time_step_min (
float
, optional, defaults to 0.001) — Minimumtime_step
used to bounddt_proj_bias
. - time_step_max (
float
, optional, defaults to 0.1) — Maximumtime_step
used to bounddt_proj_bias
. - time_step_floor (
float
, optional, defaults to 0.0001) — Minimum clamping value of thedt_proj.bias
layer initialization. - mamba_conv_bias (
bool
, optional, defaults toTrue
) — Flag indicating whether or not to use bias in the convolution layer of the mamba mixer block. - mamba_proj_bias (
bool
, optional, defaults toFalse
) — Flag indicating whether or not to use bias in the input and output projections ([“in_proj”, “out_proj”]) of the mamba mixer block
This is the configuration class to store the configuration of a ZambaModel. It is used to instantiate a Zamba 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 Zamba-v0.1 model.
Configuration objects inherit from PretrainedConfig and can be used to control the model outputs. Read the documentation from PretrainedConfig for more information.
ZambaModel
class transformers.ZambaModel
< source >( config: ZambaConfig )
Parameters
- config (ZambaConfig) — 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. config — ZambaConfig
The bare Zamba 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 ZambaDecoderLayer
forward
< source >( input_ids: LongTensor = None attention_mask: Optional = None position_ids: Optional = None past_key_values: Optional = None inputs_embeds: Optional = None use_cache: Optional = None output_attentions: Optional = None output_hidden_states: Optional = None return_dict: Optional = None cache_position: Optional = None )
Parameters
- input_ids (
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.
- 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.
Indices can be obtained using AutoTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.
If
past_key_values
is used, optionally only the lastinput_ids
have to be input (seepast_key_values
).If you want to change padding behavior, you should read
modeling_opt._prepare_decoder_attention_mask
and modify to your needs. See diagram 1 in the paper for more information on the default strategy.- 1 indicates the head is not masked,
- 0 indicates the head is masked.
- 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]
. - past_key_values (
HybridMambaAttentionDynamicCache
, optional, returned whenuse_cache=True
is passed or whenconfig.use_cache=True
) — A HybridMambaAttentionDynamicCache object containing pre-computed hidden-states (keys and values in the self-attention blocks and convolution and ssm states in the mamba blocks) that can be used (seepast_key_values
input) to speed up sequential decoding. Key and value cache tensors have shape(batch_size, num_heads, seq_len, head_dim)
. Convolution and ssm states tensors have shape(batch_size, d_inner, d_conv)
and(batch_size, d_inner, d_state)
respectively. See theHybridMambaAttentionDynamicCache
class for more details.If
past_key_values
are used, the user can optionally input only the lastinput_ids
(those that don’t have their past key value states given to this model) of shape(batch_size, 1)
instead of allinput_ids
of shape(batch_size, sequence_length)
. - 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 convertinput_ids
indices into associated vectors than the model’s internal embedding lookup matrix. - 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
). - 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. - cache_position (
torch.LongTensor
of shape(sequence_length)
, optional) — Indices depicting the position of the input sequence tokens in the sequence. Contrarily toposition_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 ZambaModel 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.
ZambaForCausalLM
forward
< source >( input_ids: LongTensor = None attention_mask: Optional = None position_ids: Optional = None past_key_values: Optional = None inputs_embeds: Optional = None labels: Optional = None use_cache: Optional = None output_attentions: Optional = None output_hidden_states: Optional = None return_dict: Optional = None cache_position: Optional = None num_logits_to_keep: int = 0 **loss_kwargs ) → 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. Padding will be ignored by default should you provide it.Indices can be obtained using AutoTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.
- 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.
Indices can be obtained using AutoTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.
If
past_key_values
is used, optionally only the lastinput_ids
have to be input (seepast_key_values
).If you want to change padding behavior, you should read
modeling_opt._prepare_decoder_attention_mask
and modify to your needs. See diagram 1 in the paper for more information on the default strategy.- 1 indicates the head is not masked,
- 0 indicates the head is masked.
- 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]
. - past_key_values (
HybridMambaAttentionDynamicCache
, optional, returned whenuse_cache=True
is passed or whenconfig.use_cache=True
) — A HybridMambaAttentionDynamicCache object containing pre-computed hidden-states (keys and values in the self-attention blocks and convolution and ssm states in the mamba blocks) that can be used (seepast_key_values
input) to speed up sequential decoding. Key and value cache tensors have shape(batch_size, num_heads, seq_len, head_dim)
. Convolution and ssm states tensors have shape(batch_size, d_inner, d_conv)
and(batch_size, d_inner, d_state)
respectively. See theHybridMambaAttentionDynamicCache
class for more details.If
past_key_values
are used, the user can optionally input only the lastinput_ids
(those that don’t have their past key value states given to this model) of shape(batch_size, 1)
instead of allinput_ids
of shape(batch_size, sequence_length)
. - 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 convertinput_ids
indices into associated vectors than the model’s internal embedding lookup matrix. - 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
). - 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. - cache_position (
torch.LongTensor
of shape(sequence_length)
, optional) — Indices depicting the position of the input sequence tokens in the sequence. Contrarily toposition_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 (seeinput_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]
.num_logits_to_keep (
int
orNone
, optional): Calculate logits for the lastnum_logits_to_keep
tokens. IfNone
, calculate logits for allinput_ids
. 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.
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 (ZambaConfig) 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 ZambaForCausalLM 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, ZambaForCausalLM
>>> model = ZambaForCausalLM.from_pretrained("Zyphra/Zamba-7B-v1")
>>> tokenizer = AutoTokenizer.from_pretrained("Zyphra/Zamba-7B-v1")
>>> prompt = "Hey, are you conscious? Can you talk to me?"
>>> 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]
"Hey, are you conscious? Can you talk to me?\nI'm not conscious, but I can talk to you."
ZambaForSequenceClassification
class transformers.ZambaForSequenceClassification
< source >( config )
Parameters
- config (ZambaConfig) — 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 Zamba Model with a sequence classification head on top (linear layer).
ZambaForSequenceClassification uses the last token in order to do the classification, as other causal models (e.g. GPT-2) do.
Since it does classification on the last token, it requires to know the position of the last token. If a
pad_token_id
is defined in the configuration, it finds the last token that is not a padding token in each row. If
no pad_token_id
is defined, it simply takes the last value in each row of the batch. Since it cannot guess the
padding tokens when inputs_embeds
are passed instead of input_ids
, it does the same (take the last value in
each row of the batch).
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.
forward
< source >( input_ids: Optional = None attention_mask: Optional = None position_ids: Optional = None past_key_values: Union = None inputs_embeds: Optional = None labels: Optional = None use_cache: Optional = None output_attentions: Optional = None output_hidden_states: Optional = None return_dict: Optional = None )
Parameters
- input_ids (
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.
- 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.
Indices can be obtained using AutoTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.
If
past_key_values
is used, optionally only the lastinput_ids
have to be input (seepast_key_values
).If you want to change padding behavior, you should read
modeling_opt._prepare_decoder_attention_mask
and modify to your needs. See diagram 1 in the paper for more information on the default strategy.- 1 indicates the head is not masked,
- 0 indicates the head is masked.
- 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]
. - past_key_values (
HybridMambaAttentionDynamicCache
, optional, returned whenuse_cache=True
is passed or whenconfig.use_cache=True
) — A HybridMambaAttentionDynamicCache object containing pre-computed hidden-states (keys and values in the self-attention blocks and convolution and ssm states in the mamba blocks) that can be used (seepast_key_values
input) to speed up sequential decoding. Key and value cache tensors have shape(batch_size, num_heads, seq_len, head_dim)
. Convolution and ssm states tensors have shape(batch_size, d_inner, d_conv)
and(batch_size, d_inner, d_state)
respectively. See theHybridMambaAttentionDynamicCache
class for more details.If
past_key_values
are used, the user can optionally input only the lastinput_ids
(those that don’t have their past key value states given to this model) of shape(batch_size, 1)
instead of allinput_ids
of shape(batch_size, sequence_length)
. - 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 convertinput_ids
indices into associated vectors than the model’s internal embedding lookup matrix. - 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
). - 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. - cache_position (
torch.LongTensor
of shape(sequence_length)
, optional) — Indices depicting the position of the input sequence tokens in the sequence. Contrarily toposition_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. - 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]
. Ifconfig.num_labels == 1
a regression loss is computed (Mean-Square loss), Ifconfig.num_labels > 1
a classification loss is computed (Cross-Entropy).
The ZambaForSequenceClassification 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.