콘텐츠로 이동

Stable Diffusion 3용 멀티모달 확산 트랜스포머(MMDiT)

RBLNSD3Transformer2DModel은 Stable Diffusion 3 모델에서 사용되는 핵심 트랜스포머 블록의 RBLN 최적화 버전입니다.

지원 NPU

Diffusers 파이프라인은 ATOM™에서만 실행됩니다. Rebel100™은 아직 지원하지 않습니다.

이 모델은 이전 Stable Diffusion 버전에서 사용된 UNet 아키텍처를 대체합니다. 여러 텍스트 인코더의 풀링된 임베딩 및 타임스텝 정보와 함께 잠재 이미지 표현을 처리하여 확산 과정을 처리합니다.

파이프라인 내 사용법

일반적으로 RBLNSD3Transformer2DModel과 직접 상호 작용하지 않습니다. 대신, RBLNStableDiffusion3Pipeline과 같은 RBLN Stable Diffusion 3 파이프라인의 일부로 자동 로드 및 관리됩니다.

RBLN SD3 파이프라인을 구성할 때 파이프라인 설정 객체의 transformer 인수를 통해 트랜스포머에 대한 특정 설정을 전달할 수 있습니다:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
from optimum.rbln import RBLNStableDiffusion3PipelineConfig

# 예시: 트랜스포머에 특정 배치 크기를 사용하여 파이프라인 구성
config = RBLNStableDiffusion3PipelineConfig(
    batch_size=1,  # 파이프라인 추론 배치 크기
    height=1024,
    width=1024,
    transformer=dict(
        batch_size=2  # 트랜스포머 배치 크기 (예: CFG를 위해 2배)
    ),
)

# ... 이 설정을 사용하여 파이프라인 로드 ...

guidance scale이 트랜스포머 배치 크기에 주는 영향을 처리하는 방법을 비롯해 파이프라인 사용·구성의 자세한 내용은 Stable Diffusion 3 파이프라인 문서에서 확인할 수 있습니다.

API 참조

RBLNSD3Transformer2DModel

Bases: RBLNModel

RBLN implementation of SD3Transformer2DModel for diffusion models like Stable Diffusion 3.

The SD3Transformer2DModel takes text and/or image embeddings from encoders (like CLIP) and maps them to a shared latent space that guides the diffusion process to generate the desired image.

This class inherits from [RBLNModel]. Check the superclass documentation for the generic methods the library implements for all its models.

forward(hidden_states, encoder_hidden_states=None, pooled_projections=None, timestep=None, block_controlnet_hidden_states=None, joint_attention_kwargs=None, return_dict=True, **kwargs)

Forward pass for the RBLN-optimized SD3Transformer2DModel.

Parameters:

Name Type Description Default
hidden_states FloatTensor

The currently predicted image embeddings.

 required
encoder_hidden_states FloatTensor

Conditional embeddings (embeddings computed from the input conditions such as prompts) to use.

 None
pooled_projections FloatTensor

Embeddings projected from the embeddings of input conditions.

 None
timestep LongTensor

Current denoising step.

 None
return_dict bool

Whether or not to return a [~diffusers.models.modeling_output.Transformer2DModelOutput] instead of a plain tuple.

 True

Returns:

Type Description
Transformer2DModelOutput | tuple

(~diffusers.models.modeling_output.Transformer2DModelOutput | tuple)

from_pretrained(model_id, export=None, rbln_config=None, **kwargs) classmethod

The from_pretrained() function is utilized in its standard form as in the HuggingFace transformers library. User can use this function to load a pre-trained model from the HuggingFace library and convert it to a RBLN model to be run on RBLN NPUs.

Parameters:

Name Type Description Default
model_id str | Path

The model id of the pre-trained model to be loaded. It can be downloaded from the HuggingFace model hub or a local path, or a model id of a compiled model using the RBLN Compiler.

 required
export bool | None

A boolean flag to indicate whether the model should be compiled. If None, it will be determined based on the existence of the compiled model files in the model_id.

 None
rbln_config dict | RBLNModelConfig | None

Configuration for RBLN model compilation and runtime. This can be provided as a dictionary or an instance of the model's configuration class (e.g., RBLNLlamaForCausalLMConfig for Llama models). For detailed configuration options, see the specific model's configuration class documentation.

 None
kwargs Any

Additional keyword arguments. Arguments with the prefix rbln_ are passed to rbln_config, while the remaining arguments are passed to the HuggingFace library.

 {}

Returns:

Type Description
RBLNModel

A RBLN model instance ready for inference on RBLN NPU devices.

get_hf_class() classmethod

Lazily load and cache the HuggingFace class this model wraps.

The class is looked up in hf_library_name under the name without the RBLN prefix (e.g. RBLNLlamaForCausalLM -> LlamaForCausalLM).

save_pretrained(save_directory, push_to_hub=False, **kwargs)

Saves a model and its configuration file to a directory, so that it can be re-loaded using the [~optimum.rbln.modeling_base.RBLNBaseModel.from_pretrained] class method.

Parameters:

Name Type Description Default
save_directory str | Path

Directory where to save the model file.

 required
push_to_hub bool

Whether or not to push your model to the HuggingFace model hub after saving it.

 False

from_model(model, config=None, rbln_config=None, model_save_dir=None, subfolder='', **kwargs) classmethod

Converts and compiles a pre-trained HuggingFace library model into a RBLN model. This method performs the actual model conversion and compilation process.

Parameters:

Name Type Description Default
model PreTrainedModel | ModelMixin

The PyTorch model to be compiled. The object must be a HuggingFace transformers PreTrainedModel or a Diffusers component based on diffusers.ModelMixin, such as a VAE, UNet or ControlNet.

 required
config PretrainedConfig | None

The configuration object associated with the model.

 None
rbln_config RBLNModelConfig | dict | None

Configuration for RBLN model compilation and runtime. This can be provided as a dictionary or an instance of the model's configuration class (e.g., RBLNLlamaForCausalLMConfig for Llama models). For detailed configuration options, see the specific model's configuration class documentation.

 None
kwargs Any

Additional keyword arguments. Arguments with the prefix rbln_ are passed to rbln_config, while the remaining arguments are passed to the HuggingFace library.

 {}

The method performs the following steps:

  1. Compiles the PyTorch model into an optimized RBLN graph
  2. Configures the model for the specified NPU device
  3. Creates the necessary runtime objects if requested
  4. Saves the compiled model and configurations

Returns:

Type Description
RBLNModel

A RBLN model instance ready for inference on RBLN NPU devices.

RBLNSD3Transformer2DModelConfig

Bases: RBLNModelConfig

Configuration class for RBLN Stable Diffusion 3 Transformer models.

This class inherits from RBLNModelConfig and provides specific configuration options for Transformer models used in diffusion models like Stable Diffusion 3.

__init__(batch_size=None, sample_size=None, prompt_embed_length=None, **kwargs)

Parameters:

Name Type Description Default
batch_size int | None

The batch size for inference. Defaults to 1.

 None
sample_size int | tuple[int, int] | None

The spatial dimensions (height, width) of the generated samples. If an integer is provided, it's used for both height and width.

 None
prompt_embed_length int | None

The length of the embedded prompt vectors that will be used to condition the transformer model.

 None
kwargs Any

Additional arguments passed to the parent RBLNModelConfig.

 {}

Raises:

Type Description
ValueError

If batch_size is not a positive integer.

from_pretrained(path, rbln_config=None, return_unused_kwargs=False, **kwargs) classmethod

Load a RBLNModelConfig from a path.

Parameters:

Name Type Description Default
path str

Path to the RBLNModelConfig file or directory containing the config file.

 required
rbln_config dict[str, Any] | None

Additional configuration to override.

 None
return_unused_kwargs bool

Whether to return unused kwargs.

 False
kwargs Any

Additional keyword arguments to override configuration values. Keys starting with 'rbln_' will have the prefix removed and be used to update the configuration.

 {}

Returns:

Name Type Description
RBLNModelConfig Self | tuple[Self, dict[str, Any]]

The loaded configuration instance.
Note

This method loads the configuration from the specified path and applies any provided overrides. If the loaded configuration class doesn't match the expected class, a warning will be logged.

Examples:

1
config = RBLNResNetForImageClassificationConfig.from_pretrained("/path/to/model")

initialize_submodule_config(submodule_config=None, force_kwargs=False, **kwargs)

Build the config of a submodule from the parent's runtime options and kwargs.

Returns the submodule config instance when its class is known (an instance was passed, or cls_name was given), otherwise the merged kwargs dict for the model class to resolve later.

get_load_overrides()

Extract the options that must cross the load boundary, for passing to from_pretrained.

Returns the explicitly-set runtime options (device, device_map, create_runtimes, activate_profiler, timeout) and the subclass-specific non-save attributes (load-behavior flags such as _load_visual_runtime, which are never serialized so they can only travel with the caller), recursively including submodules. Compile-time attributes are not included: when loading a compiled model they come from the artifact's rbln_config.json, and this dict is merged on top of it.

Returns:

Type Description
dict[str, Any]

Dictionary of load overrides, keyed like the rbln_config dict accepted by from_pretrained.