Hugging Face's logo

Transformers documentation

DBRX

Hugging Face's logo
Join the Hugging Face community

and get access to the augmented documentation experience

to get started

DBRX

개요

DBRX는 트랜스포머 기반의 다음 토큰을 예측하는 디코더 전용 LLM 모델입니다. 총 132B 매개변수를 가진 세밀한 전문가 혼합(MoE) 아키텍처를 사용하며, 이 중 36B 매개변수가 입력마다 활성화됩니다. 12T 토큰의 텍스트와 코드 데이터로 사전 학습되었습니다.

Mixtral-8x7B와 Grok-1과 같은 다른 공개 MoE 모델들과 비교했을 때, DBRX는 더 많은 수의 작은 전문가들을 사용하는 세밀한 구조를 가지고 있습니다. DBRX는 16개의 전문가 중 4개를 선택하는 반면, Mixtral-8x7B와 Grok-1은 8개의 전문가 중 2개를 선택합니다.

이는 65배 더 많은 전문가 조합을 가능하게 하며, 이를 통해 모델의 품질이 향상되는 것을 발견했습니다. DBRX는 회전 위치 인코딩(RoPE), 게이트 선형 유닛(GLU), 그룹 쿼리 어텐션(GQA)을 사용합니다. BPE 기반 모델이며 tiktoken 저장소에 설명된 GPT-4 토크나이저를 사용합니다. 이러한 선택들은 철저한 평가와 스케일링 실험을 기반으로 이루어졌습니다.

DBRX는 신중하게 선별된 12T 토큰의 데이터로 사전 학습되었으며, 최대 문맥 길이는 32K 토큰입니다. 이 데이터는 토큰 대비 MPT 계열 모델 학습에 사용된 데이터보다 최소 2배 이상 더 좋은 것으로 추정됩니다. 이 새로운 데이터셋은 데이터 처리를 위한 Apache Spark™와 Databricks 노트북, 그리고 데이터 관리와 거버넌스를 위한 Unity Catalog를 포함한 Databricks 도구 전체를 활용하여 개발되었습니다. 우리는 사전 학습을 위해 커리큘럼 학습을 사용했으며, 학습 중 데이터 믹스를 변경하는 방식이 모델 품질을 상당히 개선한다는 것을 발견했습니다.

DBRX Instruct와 DBRX Base에 대한 더 자세한 정보는 이 기술 블로그 포스트에서 확인할 수 있습니다.

이 모델은 eitan-turok와 abhi-db가 기여했습니다. 원본 코드는 이곳에서 찾을 수 있지만, 최신 버전이 아닐 수 있습니다.

사용 예

generate() 메소드는 DBRX를 사용하여 텍스트를 생성하는 데 사용될 수 있습니다. 표준 어텐션 구현, 플래시 어텐션, PyTorch의 스케일된 내적 어텐션(Scaled Dot-Product Attention)을 사용하여 생성할 수 있습니다. 후자의 두 어텐션 구현 방식은 처리 속도를 크게 높여줍니다.

from transformers import DbrxForCausalLM, AutoTokenizer
import torch

tokenizer = AutoTokenizer.from_pretrained("databricks/dbrx-instruct", token="YOUR_HF_TOKEN")
model = DbrxForCausalLM.from_pretrained(
    "databricks/dbrx-instruct",
    device_map="auto",
    torch_dtype=torch.bfloat16,
    token="YOUR_HF_TOKEN",
    )

input_text = "What does it take to build a great LLM?"
messages = [{"role": "user", "content": input_text}]
input_ids = tokenizer.apply_chat_template(messages, return_dict=True, tokenize=True, add_generation_prompt=True, return_tensors="pt").to("cuda")

outputs = model.generate(**input_ids, max_new_tokens=200)
print(tokenizer.decode(outputs[0]))

pip install flash-attn를 통해 플래시 어텐션을 설치하면, 더 빠른 생성이 가능합니다. (플래시 어텐션에 대한 HuggingFace 문서는 이곳에서 확인할 수 있습니다.)

from transformers import DbrxForCausalLM, AutoTokenizer
import torch

tokenizer = AutoTokenizer.from_pretrained("databricks/dbrx-instruct", token="YOUR_HF_TOKEN")
model = DbrxForCausalLM.from_pretrained(
    "databricks/dbrx-instruct",
    device_map="auto",
    torch_dtype=torch.bfloat16,
    token="YOUR_HF_TOKEN",
    attn_implementation="flash_attention_2",
    )

input_text = "What does it take to build a great LLM?"
messages = [{"role": "user", "content": input_text}]
input_ids = tokenizer.apply_chat_template(messages, return_dict=True, tokenize=True, add_generation_prompt=True, return_tensors="pt").to("cuda")

outputs = model.generate(**input_ids, max_new_tokens=200)
print(tokenizer.decode(outputs[0]))

PyTorch의 스케일된 내적 어텐션을 사용하여도 더 빠른 생성이 가능합니다. (스케일된 내적 어텐션에 대한 HuggingFace 문서는 이곳에서 확인할 수 있습니다.)

from transformers import DbrxForCausalLM, AutoTokenizer
import torch

tokenizer = AutoTokenizer.from_pretrained("databricks/dbrx-instruct", token="YOUR_HF_TOKEN")
model = DbrxForCausalLM.from_pretrained(
    "databricks/dbrx-instruct",
    device_map="auto",
    torch_dtype=torch.bfloat16,
    token="YOUR_HF_TOKEN",
    attn_implementation="sdpa",
    )

input_text = "What does it take to build a great LLM?"
messages = [{"role": "user", "content": input_text}]
input_ids = tokenizer.apply_chat_template(messages, return_dict=True, tokenize=True, add_generation_prompt=True, return_tensors="pt").to("cuda")

outputs = model.generate(**input_ids, max_new_tokens=200)
print(tokenizer.decode(outputs[0]))

DbrxConfig

class transformers.DbrxConfig

< >

( d_model: int = 2048 n_heads: int = 16 n_layers: int = 24 max_seq_len: int = 2048 vocab_size: int = 32000 resid_pdrop: float = 0.0 emb_pdrop: float = 0.0 attn_config: typing.Optional[transformers.models.dbrx.configuration_dbrx.DbrxAttentionConfig] = None ffn_config: typing.Optional[transformers.models.dbrx.configuration_dbrx.DbrxFFNConfig] = None use_cache: bool = True initializer_range: float = 0.02 output_router_logits: bool = False **kwargs: typing.Any )

Parameters

  • d_model (int, optional, defaults to 2048) — Dimensionality of the embeddings and hidden states.
  • n_heads (int, optional, defaults to 16) — Number of attention heads for each attention layer in the Transformer encoder.
  • n_layers (int, optional, defaults to 24) — Number of hidden layers in the Transformer encoder.
  • max_seq_len (int, optional, defaults to 2048) — The maximum sequence length of the model.
  • vocab_size (int, optional, defaults to 32000) — Vocabulary size of the Dbrx model. Defines the maximum number of different tokens that can be represented by the inputs_ids passed when calling DbrxModel.
  • resid_pdrop (float, optional, defaults to 0.0) — The dropout probability applied to the attention output before combining with residual.
  • emb_pdrop (float, optional, defaults to 0.0) — The dropout probability for the embedding layer.
  • attn_config (dict, optional) — A dictionary used to configure the model’s attention module.
  • ffn_config (dict, optional) — A dictionary used to configure the model’s FFN module.
  • use_cache (bool, optional, defaults to True) — Whether or not the model should return the last key/values attentions (not used by all models).
  • initializer_range (float, optional, defaults to 0.02) — The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
  • output_router_logits (bool, optional, defaults to False) — Whether or not the router logits should be returned by the model. Enabling this will also allow the model to output the auxiliary loss. See here for more details.

This is the configuration class to store the configuration of a DbrxModel. It is used to instantiate a Dbrx model according to the specified arguments, defining the model architecture. Instantiating a configuration with the defaults will yield a different configuration to that of the databricks/dbrx-instruct architecture.

Configuration objects inherit from PretrainedConfig and can be used to control the model outputs. Read the documentation from PretrainedConfig for more information.

Example:

>>> from transformers import DbrxConfig, DbrxModel

>>> # Initializing a Dbrx configuration
>>> configuration = DbrxConfig(n_layers=2, d_model=256, n_heads=8, vocab_size=128)

>>> # Initializing a model (with random weights) from the configuration
>>> model = DbrxModel(configuration)

>>> # Accessing the model configuration
>>> configuration = model.config

DbrxModel

class transformers.DbrxModel

< >

( config: DbrxConfig )

Parameters

  • config (DbrxConfig) — 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 (DbrxConfig) — Model configuration class with all 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 DBRX 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. Each layer is a DbrxBlock layer.

forward

< >

( input_ids: typing.Optional[torch.LongTensor] = None attention_mask: typing.Optional[torch.Tensor] = None position_ids: typing.Optional[torch.LongTensor] = None past_key_values: typing.Optional[transformers.cache_utils.Cache] = None inputs_embeds: typing.Optional[torch.Tensor] = None use_cache: typing.Optional[bool] = None output_attentions: typing.Optional[bool] = None output_hidden_states: typing.Optional[bool] = None output_router_logits: typing.Optional[bool] = None return_dict: typing.Optional[bool] = None cache_position: typing.Optional[torch.LongTensor] = 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.

    What are input 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?

    Indices can be obtained using AutoTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.

    If past_key_values is used, optionally only the last decoder_input_ids have to be input (see past_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].

    What are position IDs?

  • past_key_values (Cache or tuple(tuple(torch.FloatTensor)), 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.

    Two formats are allowed:

    • a Cache instance, see our kv cache guide;
    • 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)). This is also known as the legacy cache format.

    The model will output the same cache format that is fed as input. If no past_key_values are passed, the legacy cache format will be returned.

    If past_key_values are used, the user can optionally input only the last input_ids (those that don’t have their past key value states given to this model) of shape (batch_size, 1) instead of all input_ids of shape (batch_size, sequence_length).

  • 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_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.
  • output_router_logits (bool, optional) — Whether or not to return the logits of all the routers. They are useful for computing the router loss, and should not be returned during inference.
  • 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 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 DbrxModel 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.

DbrxForCausalLM

class transformers.DbrxForCausalLM

< >

( config: DbrxConfig )

Parameters

  • config (DbrxConfig) — 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 DBRX Model transformer for causal language modeling. 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

< >

( input_ids: typing.Optional[torch.LongTensor] = None attention_mask: typing.Optional[torch.Tensor] = None position_ids: typing.Optional[torch.LongTensor] = None past_key_values: typing.Optional[transformers.cache_utils.Cache] = None inputs_embeds: typing.Optional[torch.Tensor] = None labels: typing.Optional[torch.LongTensor] = None use_cache: typing.Optional[bool] = None output_attentions: typing.Optional[bool] = None output_hidden_states: typing.Optional[bool] = None output_router_logits: typing.Optional[bool] = None return_dict: typing.Optional[bool] = None cache_position: typing.Optional[torch.LongTensor] = None num_logits_to_keep: int = 0 ) → transformers.modeling_outputs.MoeCausalLMOutputWithPast 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.

Returns

transformers.modeling_outputs.MoeCausalLMOutputWithPast or tuple(torch.FloatTensor)

A transformers.modeling_outputs.MoeCausalLMOutputWithPast 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 (DbrxConfig) 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).

  • aux_loss (torch.FloatTensor, optional, returned when labels is provided) — aux_loss for the sparse modules.

  • router_logits (tuple(torch.FloatTensor), optional, returned when output_router_probs=True and config.add_router_probs=True is passed or when config.output_router_probs=True) — Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, sequence_length, num_experts).

    Raw router logtis (post-softmax) that are computed by MoE routers, these terms are used to compute the auxiliary loss for Mixture of Experts models.

  • 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))

    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 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 DbrxForCausalLM 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.

Forward function for causal language modeling.

Example:

>> from transformers import AutoTokenizer, DbrxForCausalLM

>> model = DbrxForCausalLM.from_pretrained("databricks/dbrx-instruct")
>> tokenizer = AutoTokenizer.from_pretrained("databricks/dbrx-instruct")

>> 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."
< > Update on GitHub

←(번역중) CTRL DeBERTa→