Hugging Face's logo

Transformers documentation

Muse Spark Open

You are viewing main version, which requires installation from source. If you'd like regular pip install, checkout the latest stable version (v5.20.0).
Hugging Face's logo
Join the Hugging Face community

and get access to the augmented documentation experience

to get started

This model was contributed to Hugging Face Transformers on 2026-10-14.

FlashAttention SDPA

Muse Spark Open

Overview

Muse Spark Open is a multimodal mixture-of-experts (MoE) model developed by Meta that accepts text, images, audio, and video as input and generates text. It has 816 billion total parameters, with 42 billion active per token, and supports a context window of up to 1,048,576 tokens shared across all input modalities. The model is trained for agentic coding tasks, including multi-file code editing, debugging, and tool use.

The language backbone consists of 62 transformer layers, each routing tokens to 8 of 256 experts. It uses grouped-query attention with query-key normalization, attention output gating and normalization, and learned residual gates. A shared perception encoder processes images and video frames, while a convolutional audio encoder produces audio representations at 12.5 tokens per second. Muse Spark Open shares its tokenizer and vocabulary with Muse Glimmer, with extensions to the chat template for multimodal inputs. Checkpoints are available in BF16 and NVFP4 formats.

Resources

AutoModel
AutoModel Image
AutoModel Video
AutoModel Audio
import os

from transformers import AutoModelForMultimodalLM, AutoProcessor, DistributedConfig

distributed_config = DistributedConfig(
    tp_size=int(os.environ["WORLD_SIZE"]), # Number of devices across which the dense layers are sharded
    ep_size=int(os.environ["WORLD_SIZE"]), # Number of devices across which the experts are distributed
)

processor = AutoProcessor.from_pretrained("meta-models/Muse-Spark-Open-816B-A42B")
model = AutoModelForMultimodalLM.from_pretrained(
    "meta-models/Muse-Spark-Open-816B-A42B",
    distributed_config=distributed_config,
)

messages = [
    {"role": "user", "content": "Explain in one sentence why the sky is usually blue."},
]
inputs = processor.apply_chat_template(
    messages,
    reasoning_effort="low",  # default: xhigh, options: low/medium/high/xhigh
    add_generation_prompt=True,
    tokenize=True,
    return_dict=True,
    return_tensors="pt",
).to(model.device).to(model.dtype)
input_len = inputs["input_ids"].shape[-1]

outputs = model.generate(**inputs, max_new_tokens=512)
print(processor.decode(outputs[0][input_len:], skip_special_tokens=False))
#  to=self<|message|>Explain in one sentence why the sky is usually blue.
# ...
# <|eom|><|start|>assistant to=user<|message|>The sky is usually blue because molecules in Earth's atmosphere scatter shorter-wavelength blue sunlight in all directions much more strongly than longer-wavelength red light.<|eot|>

NVFP4 Weights

NVFP4 quantized weights with lower memory requirements are available on the hub. The model can be loaded on any NVFP4 compatible device without changes to the code. The quantization configuration is automatically applied at load time.

For NVFP4 support, transformers must be installed from this branch.

pip install git+https://github.com/huggingface/transformers.git@finegrained-per-scheme-demo
import os

from transformers import AutoModelForMultimodalLM, AutoProcessor, DistributedConfig

distributed_config = DistributedConfig(
    tp_size=int(os.environ["WORLD_SIZE"]),
    ep_size=int(os.environ["WORLD_SIZE"]),
)

processor = AutoProcessor.from_pretrained("meta-models/Muse-Spark-Open-816B-A42B-NVFP4")
model = AutoModelForMultimodalLM.from_pretrained(
    "meta-models/Muse-Spark-Open-816B-A42B-NVFP4",
    distributed_config=distributed_config,
)

# ...

MuseSparkConfig

class transformers.MuseSparkConfig

< >

( transformers_version: str | None = Nonearchitectures: list[str] | None = Noneoutput_hidden_states: bool | None = Falsereturn_dict: bool | None = Truedtype: str | torch.dtype | None = Nonechunk_size_feed_forward: int = 0is_encoder_decoder: bool = Falseid2label: dict[int, str] | dict[str, str] | None = Nonelabel2id: dict[str, int] | dict[str, str] | None = Noneproblem_type: Literal['regression', 'single_label_classification', 'multi_label_classification'] | None = Nonetext_config: dict | transformers.configuration_utils.PreTrainedConfig | None = Nonevision_config: dict | transformers.configuration_utils.PreTrainedConfig | None = Noneimage_token_id: int = 200092video_token_id: int = 200091out_hidden_size: int = 6144projector_hidden_size: int = 4096projector_hidden_act: str = 'gelu'audio_config: dict | transformers.configuration_utils.PreTrainedConfig | None = Noneaudio_token_id: int = 200038 )

Parameters

  • text_config (Union[dict, ~configuration_utils.PreTrainedConfig], optional) — The config object or dictionary of the text backbone.
  • vision_config (Union[dict, ~configuration_utils.PreTrainedConfig], optional) — The config object or dictionary of the vision backbone.
  • image_token_id (int, optional, defaults to 200092) — The image token index used as a placeholder for input images.
  • video_token_id (int, optional, defaults to 200091) — The video token index used as a placeholder for input videos.
  • out_hidden_size (int, optional, defaults to 6144) — Output dimension of the vision encoder after patch merging (input width of the multimodal projection).
  • projector_hidden_size (int, optional, defaults to 4096) — Intermediate dimension of the multimodal projection.
  • projector_hidden_act (str, optional, defaults to gelu) — The activation function used by the multimodal projector.
  • audio_config (Union[dict, ~configuration_utils.PreTrainedConfig], optional) — The config object or dictionary of the audio backbone.
  • audio_token_id (int, optional, defaults to 200038) — The audio token index used as a placeholder for input audio.

This is the configuration class to store the configuration of a MuseSparkModel. It is used to instantiate a Muse Spark 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 meta-models/Muse-Spark-Open-816B-A42B

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 MuseSparkForConditionalGeneration, MuseSparkConfig

>>> # Initializing an MuseSpark style configuration
>>> configuration = MuseSparkConfig()

>>> # Initializing a model from the configuration
>>> model = MuseSparkForConditionalGeneration(configuration)

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

MuseSparkTextConfig

class transformers.MuseSparkTextConfig

< >

( transformers_version: str | None = Nonearchitectures: list[str] | None = Noneoutput_hidden_states: bool | None = Falsereturn_dict: bool | None = Truedtype: str | torch.dtype | None = Nonechunk_size_feed_forward: int = 0is_encoder_decoder: bool = Falseid2label: dict[int, str] | dict[str, str] | None = Nonelabel2id: dict[str, int] | dict[str, str] | None = Noneproblem_type: Literal['regression', 'single_label_classification', 'multi_label_classification'] | None = Nonevocab_size: int = 202048hidden_size: int = 8192intermediate_size: int = 16384num_hidden_layers: int = 62num_attention_heads: int = 128num_key_value_heads: int = 16head_dim: int = 64max_position_embeddings: int = 1048576initializer_range: float = 0.02rms_norm_eps: float = 1e-05use_cache: bool = Truepad_token_id: int | None = Noneeos_token_id: int | list[int] | None = 200001bos_token_id: int | None = 200000tie_word_embeddings: bool = Falserope_parameters: transformers.modeling_rope_utils.RopeParameters | dict | None = Noneattention_bias: bool = Falseattention_dropout: int | float | None = 0.0sliding_window: int | None = 2048layer_types: list[str] | None = Nonefinal_logit_softcapping: float = 20.0qk_scale_factor: float = 5.4730064863875output_multiplier: float = 0.1767766952966369post_norm_eps: float = 1e-08layer_rope_theta: list[float | int] | None = Nonehidden_act: str = 'silu'moe_intermediate_size: int = 4096num_local_experts: int = 256num_experts_per_tok: int = 8routed_scaling_factor: float = 1.0moe_hidden_size: int = 4096mlp_layer_types: list[str] | None = Noneresidual_gate_temperature: float = 0.3output_router_logits: bool = False )

Parameters

  • vocab_size (int, optional, defaults to 202048) — 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 8192) — Dimension of the hidden representations.
  • intermediate_size (int, optional, defaults to 16384) — Dimension of the MLP representations.
  • num_hidden_layers (int, optional, defaults to 62) — Number of hidden layers in the Transformer decoder.
  • num_attention_heads (int, optional, defaults to 128) — Number of attention heads for each attention layer 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. 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. If it is not specified, will default to num_attention_heads.
  • head_dim (int, optional, defaults to 64) — The attention head dimension. If None, it will default to hidden_size // num_attention_heads
  • max_position_embeddings (int, optional, defaults to 1048576) — The maximum sequence length that this model might ever be used with.
  • 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 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) — Token id used for padding in the vocabulary.
  • eos_token_id (Union[int, list[int]], optional, defaults to 200001) — Token id used for end-of-stream in the vocabulary.
  • bos_token_id (int, optional, defaults to 200000) — Token id used for beginning-of-stream in the vocabulary.
  • tie_word_embeddings (bool, optional, defaults to False) — Whether to tie weight embeddings according to model’s tied_weights_keys mapping.
  • 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.
  • attention_bias (bool, optional, defaults to False) — Whether to use a bias in the query, key, value and output projection layers during self-attention.
  • attention_dropout (Union[int, float], optional, defaults to 0.0) — The dropout ratio for the attention probabilities.
  • sliding_window (int, optional, defaults to 2048) — Sliding window attention window size. If None, no sliding window is applied.
  • layer_types (list[str], optional) — A list that explicitly maps each layer index with its layer type. If not provided, it will be automatically generated based on config values.
  • final_logit_softcapping (float, optional, defaults to 20.0) — Soft-capping value applied to the final logits before computing the probability distribution. Logits are scaled by tanh(logit / cap) * cap.
  • qk_scale_factor (float, optional, defaults to 5.4730064863875) — Multiplier applied to Q after the scaleless QK-norm, on top of the standard 1/sqrt(head_dim) attention scaling.
  • output_multiplier (float, optional, defaults to 0.1767766952966369) — Scale applied to logits before the final tanh softcap. Equal to 1/sqrt(hidden_size / 256) for the released checkpoint.
  • post_norm_eps (float, optional, defaults to 1e-8) — Epsilon used for the post-attention and post-FFN norms (which sit between the sub-layer output and the residual).
  • layer_rope_theta (list[float], optional) — Per-layer RoPE base theta; 0 disables rotary (NoPE) for that layer. Overrides the global rope_parameters["rope_theta"]. Defaults to the global theta everywhere except every 4th layer counted backward from the last, which is NoPE.
  • hidden_act (str, optional, defaults to silu) — The non-linear activation function (function or string) in the decoder. For example, "gelu", "relu", "silu", etc.
  • moe_intermediate_size (int, optional, defaults to 4096) — Intermediate size of the routed expert MLPs.
  • num_local_experts (int, optional, defaults to 256) — Number of local experts on each device. num_experts should be divisible by num_local_experts.
  • num_experts_per_tok (int, optional, defaults to 8) — Number of experts to route each token to. This is the top-k value for the token-choice routing.
  • routed_scaling_factor (float, optional, defaults to 1.0) — Scaling factor or routed experts.
  • moe_hidden_size (int, optional, defaults to 4096) — Width the MoE block pools hidden_size down to before the routed experts, and unpools from afterwards. The experts run entirely at this width.
  • mlp_layer_types (list[str], optional) — Per-layer MLP type — "dense" or "sparse". Length must equal num_hidden_layers. Defaults to all sparse.
  • residual_gate_temperature (float, optional, defaults to 0.3) — Temperature applied to the residual gate parameter before the sigmoid.
  • 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, including load balancing loss and router z-loss.

This is the configuration class to store the configuration of a MuseSparkModel. It is used to instantiate a Muse Spark 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 meta-models/Muse-Spark-Open-816B-A42B

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

MuseSparkVisionConfig

class transformers.MuseSparkVisionConfig

< >

( transformers_version: str | None = Nonearchitectures: list[str] | None = Noneoutput_hidden_states: bool | None = Falsereturn_dict: bool | None = Truedtype: str | torch.dtype | None = Nonechunk_size_feed_forward: int = 0is_encoder_decoder: bool = Falseid2label: dict[int, str] | dict[str, str] | None = Nonelabel2id: dict[str, int] | dict[str, str] | None = Noneproblem_type: Literal['regression', 'single_label_classification', 'multi_label_classification'] | None = Nonepatch_size: int = 14pos_emb_height: int = 32pos_emb_width: int = 32num_attention_heads: int = 16num_hidden_layers: int = 50hidden_size: int = 1536intermediate_size: int = 8960hidden_act: str = 'gelu'rope_parameters: dict | None = Nonepatch_temporal: int = 2merge_size: int = 2max_position_embeddings: int = 1024layer_norm_eps: float = 1e-05layer_types: list[str] | None = None )

Parameters

  • patch_size (int, optional, defaults to 14) — The size (resolution) of each patch.
  • pos_emb_height (int, optional) — Initial position embedding height.
  • pos_emb_width (int, optional) — Initial position embedding width.
  • num_attention_heads (int, optional, defaults to 16) — Number of attention heads for each attention layer in the Transformer decoder.
  • num_hidden_layers (int, optional, defaults to 50) — Number of hidden layers in the Transformer decoder.
  • hidden_size (int, optional, defaults to 1536) — Dimension of the hidden representations.
  • intermediate_size (int, optional, defaults to 8960) — Dimension of the MLP representations.
  • hidden_act (str, optional, defaults to gelu) — The non-linear activation function (function or string) in the decoder. For example, "gelu", "relu", "silu", etc.
  • rope_parameters (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.
  • patch_temporal (int, optional) — The temporal patch size used to embed inputs.
  • merge_size (tuple[int] | list[int], optional) — Kernel size for patch merging.
  • max_position_embeddings (int, optional, defaults to 1024) — The maximum sequence length that this model might ever be used with.
  • layer_norm_eps (float, optional, defaults to 1e-05) — The epsilon used by the layer normalization layers.
  • layer_types (list[str], optional) — A list that explicitly maps each layer index with its layer type. If not provided, it will be automatically generated based on config values.

This is the configuration class to store the configuration of a MuseSparkModel. It is used to instantiate a Muse Spark 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 meta-models/Muse-Spark-Open-816B-A42B

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

MuseSparkAudioConfig

class transformers.MuseSparkAudioConfig

< >

( transformers_version: str | None = Nonearchitectures: list[str] | None = Noneoutput_hidden_states: bool | None = Falsereturn_dict: bool | None = Truedtype: str | torch.dtype | None = Nonechunk_size_feed_forward: int = 0is_encoder_decoder: bool = Falseid2label: dict[int, str] | dict[str, str] | None = Nonelabel2id: dict[str, int] | dict[str, str] | None = Noneproblem_type: Literal['regression', 'single_label_classification', 'multi_label_classification'] | None = Nonepatch_sizes: list[int] | tuple[int, ...] = (16, 6, 5, 4)hidden_sizes: list[int] | tuple[int, ...] = (256, 512, 1024, 2048)depths: list[int] | tuple[int, ...] = (2, 2, 2, 2)num_hidden_layers: int = 4mlp_ratio: float = 4.0conv_kernel_size: int = 7hidden_act: str = 'gelu'rms_norm_eps: float = 1e-06out_hidden_size: int = 8192initializer_range: float = 0.02 )

Parameters

  • patch_sizes (list[int] | tuple[int, ...], optional, defaults to (16, 6, 5, 4)) — Temporal downsampling factor applied by each stage’s patch embedding. Their product is the number of waveform samples collapsed into one output token.
  • hidden_sizes (list[int] | tuple[int, ...], optional, defaults to (256, 512, 1024, 2048)) — Hidden width of each stage.
  • depths (list[int] | tuple[int, ...], optional, defaults to (2, 2, 2, 2)) — Number of conv blocks in each layer.
  • num_hidden_layers (int, optional, defaults to 4) — Number of MuseSparkAudioLayers, i.e. resolutions in the pyramid. Redundant with len(hidden_sizes) and validated against it, so a custom pyramid must set both.
  • mlp_ratio (float, optional, defaults to 4.0) — Ratio of a layer’s MLP intermediate width to its hidden width.
  • conv_kernel_size (int, optional, defaults to 7) — Kernel size of the depthwise causal convolutions.
  • hidden_act (str, optional, defaults to gelu) — The non-linear activation function (function or string) in the decoder. For example, "gelu", "relu", "silu", etc.
  • rms_norm_eps (float, optional, defaults to 1e-06) — The epsilon used by the rms normalization layers.
  • out_hidden_size (int, optional, defaults to 8192) — Width the final projection maps to, i.e. the text model’s hidden size.
  • initializer_range (float, optional, defaults to 0.02) — The standard deviation of the truncated_normal_initializer for initializing all weight matrices.

This is the configuration class to store the configuration of a MuseSparkModel. It is used to instantiate a Muse Spark 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 meta-models/Muse-Spark-Open-816B-A42B

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 MuseSparkAudioConfig, MuseSparkAudioModel

>>> configuration = MuseSparkAudioConfig()
>>> model = MuseSparkAudioModel(configuration)
>>> configuration = model.config

MuseSparkFeatureExtractor

class transformers.MuseSparkFeatureExtractor

< >

( feature_size = 1sampling_rate = 24000padding_value = 0.0samples_per_token = 1920return_attention_mask = True**kwargs )

Parameters

  • feature_size (int, optional, defaults to 1) — The feature dimension of the extracted features. Muse Spark Open takes mono audio.
  • sampling_rate (int, optional, defaults to 24000) — The sampling rate, in hertz, the audio is expected at. Audio at any other rate must be resampled before it reaches this class.
  • padding_value (float, optional, defaults to 0.0) — Value used to pad, on the right.
  • samples_per_token (int, optional, defaults to 1920) — Waveform samples the encoder collapses into one embedding — the product of the audio config’s patch_sizes (16 6 5 * 4). Batches are padded up to a multiple of this.
  • return_attention_mask (bool, optional, defaults to True) — Whether __call__() should return input_values_mask. The audio tower is fully convolutional and does not consume one; it is off by default.

Constructs a Muse Spark Open feature extractor.

Muse Spark Opens’s audio tower consumes the raw waveform — there is no spectrogram front end — so this only batches, pads and (optionally) truncates. The one non-obvious job it does is pad every batch up to a multiple of samples_per_token: the encoder collapses exactly that many samples into one embedding and raises otherwise.

This feature extractor inherits from SequenceFeatureExtractor which contains most of the main methods. Users should refer to this superclass for more information regarding those methods.

MuseSparkProcessor

class transformers.MuseSparkProcessor

< >

( image_processor = Nonevideo_processor = Nonetokenizer = Nonefeature_extractor = Nonechat_template = None**kwargs )

Parameters

  • image_processor (MuseGlimmerImageProcessor) — The image processor is a required input.
  • video_processor (MuseGlimmerVideoProcessor) — The video processor is a required input.
  • tokenizer (tokenizer_class) — The tokenizer is a required input.
  • feature_extractor (MuseSparkFeatureExtractor) — The feature extractor is a required input.
  • chat_template (str) — A Jinja template to convert lists of messages in a chat into a tokenizable string.

Constructs a MuseSparkProcessor which wraps a image processor, a video processor, a tokenizer, and a feature extractor into a single processor.

MuseSparkProcessor offers all the functionalities of MuseGlimmerImageProcessor, MuseGlimmerVideoProcessor, tokenizer_class, and MuseSparkFeatureExtractor. See the ~MuseGlimmerImageProcessor, ~MuseGlimmerVideoProcessor, ~tokenizer_class, and ~MuseSparkFeatureExtractor for more information.

MuseSparkPreTrainedModel

class transformers.MuseSparkPreTrainedModel

< >

( config: PreTrainedConfig*inputs**kwargs )

Parameters

  • config (PreTrainedConfig) — 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.

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.

MuseSparkTextModel

class transformers.MuseSparkTextModel

< >

( config: MuseSparkTextConfig )

Parameters

  • config (MuseSparkTextConfig) — 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 bare Muse Spark Text 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.

forward

< >

( input_ids: typing.Optional[torch.LongTensor] = Noneattention_mask: typing.Optional[torch.Tensor] = Noneposition_ids: typing.Optional[torch.LongTensor] = Nonepast_key_values: transformers.cache_utils.Cache | None = Noneinputs_embeds: typing.Optional[torch.FloatTensor] = Noneuse_cache: bool | None = None**kwargs: *<class 'transformers.utils.generic.TransformersKwargs'> ) → MoeModelOutputWithPast or tuple(torch.FloatTensor)

Parameters

  • 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. 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?

  • 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_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 instance is allowed as input, see our kv cache guide. If no past_key_values are passed, 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).

  • 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).

Returns

MoeModelOutputWithPast or tuple(torch.FloatTensor)

A MoeModelOutputWithPast 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 (MuseSparkConfig) and inputs.

The MuseSparkTextModel 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, sequence_length, hidden_size)) — Sequence of hidden-states at the output of the last layer of the model.

  • past_key_values (Cache, optional, returned when use_cache=True is passed or when config.use_cache=True) — It is a Cache instance. For more details, see our kv cache guide.

    Contains pre-computed hidden-states (key and values in the self-attention blocks and optionally if config.is_encoder_decoder=True in the cross-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.

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

MuseSparkVisionModel

class transformers.MuseSparkVisionModel

< >

( config: MuseSparkVisionConfig )

Parameters

  • config (MuseSparkVisionConfig) — 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 bare Muse Spark 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.

forward

< >

( pixel_values: FloatTensorgrid_thw: LongTensor**kwargs: *<class 'transformers.utils.generic.TransformersKwargs'> ) → BaseModelOutputWithPooling or tuple(torch.FloatTensor)

Parameters

Returns

BaseModelOutputWithPooling or tuple(torch.FloatTensor)

A BaseModelOutputWithPooling 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 (MuseSparkConfig) and inputs.

The MuseSparkVisionModel 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, sequence_length, hidden_size)) — Sequence of hidden-states at the output of the last layer of the model.

  • pooler_output (torch.FloatTensor of shape (batch_size, hidden_size)) — Last layer hidden-state of the first token of the sequence (classification token) after further processing through the layers used for the auxiliary pretraining task. E.g. for BERT-family of models, this returns the classification token after processing through a linear layer and a tanh activation function. The linear layer weights are trained from the next sentence prediction (classification) objective during pretraining.

  • 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.

MuseSparkAudioModel

class transformers.MuseSparkAudioModel

< >

( config: MuseSparkAudioConfig )

Parameters

  • config (MuseSparkAudioConfig) — 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 Muse Spark Open waveform audio encoder: a fully causal convolutional pyramid that maps raw mono 24 kHz audio to one embedding per 1920 samples (80 ms, i.e. 12.5 Hz), already in the text model’s hidden size. It uses no attention and no spectrogram front-end.

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_values: Tensorpadding_cache: transformers.models.muse_spark.modeling_muse_spark.MuseSparkAudioConv1dPaddingCache | None = None**kwargs: *<class 'transformers.utils.generic.TransformersKwargs'> ) → BaseModelOutput or tuple(torch.FloatTensor)

Parameters

  • input_values (torch.Tensor of shape (batch_size, sequence_length)) — Float values of input raw speech waveform. Values can be obtained by loading a .flac or .wav audio file into an array of type list[float], a numpy.ndarray or a torch.Tensor, e.g. via the torchcodec library (pip install torchcodec) or the soundfile library (pip install soundfile). To prepare the array into input_values, the AutoProcessor should be used for padding and conversion into a tensor of type torch.FloatTensor. See MuseSparkProcessor.call() for details.
  • padding_cache (MuseSparkAudioConv1dPaddingCache, optional) — Cache for padding in convolutional layers to maintain state across streaming chunks.

Returns

BaseModelOutput or tuple(torch.FloatTensor)

A BaseModelOutput 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 (MuseSparkConfig) and inputs.

The MuseSparkAudioModel 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, sequence_length, hidden_size)) — 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, 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.

MuseSparkModel

class transformers.MuseSparkModel

< >

( config: MuseSparkConfig )

Parameters

  • config (MuseSparkConfig) — 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 bare Muse Spark 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.

forward

< >

( input_ids: typing.Optional[torch.LongTensor] = Noneattention_mask: typing.Optional[torch.Tensor] = Noneposition_ids: typing.Optional[torch.LongTensor] = Nonepast_key_values: transformers.cache_utils.Cache | None = Noneinputs_embeds: typing.Optional[torch.FloatTensor] = Noneuse_cache: bool | None = Nonepixel_values: typing.Optional[torch.Tensor] = Noneimage_grid_thw: typing.Optional[torch.LongTensor] = Nonepixel_values_videos: typing.Optional[torch.Tensor] = Nonevideo_grid_thw: typing.Optional[torch.LongTensor] = Noneinput_values: typing.Optional[torch.Tensor] = Noneinput_values_mask: typing.Optional[torch.Tensor] = Nonemm_encoder_outputs: dict[str, transformers.modeling_outputs.BaseModelOutputWithPooling] | None = None**kwargs: *<class 'transformers.utils.generic.TransformersKwargs'> ) → MuseSparkModelOutputWithPast or tuple(torch.FloatTensor)

Parameters

  • 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. 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?

  • 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_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 instance is allowed as input, see our kv cache guide. If no past_key_values are passed, 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).

  • 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).
  • pixel_values (torch.Tensor of shape (batch_size, num_channels, image_size, image_size), optional) — The tensors corresponding to the input images. Pixel values can be obtained using MuseGlimmerImageProcessor. See MuseGlimmerImageProcessor.call() for details (MuseSparkProcessor uses MuseGlimmerImageProcessor for processing images).
  • image_grid_thw (torch.LongTensor of shape (num_images, 3), optional) — The temporal, height and width of feature shape of each image in LLM.
  • pixel_values_videos (torch.Tensor of shape (batch_size, num_frames, num_channels, frame_size, frame_size), optional) — The tensors corresponding to the input video. Pixel values for videos can be obtained using MuseGlimmerVideoProcessor. See MuseGlimmerVideoProcessor.call() for details (MuseSparkProcessor uses MuseGlimmerVideoProcessor for processing videos).
  • video_grid_thw (torch.LongTensor of shape (num_videos, 3), optional) — The temporal, height and width of feature shape of each video in LLM.
  • input_values (torch.Tensor of shape (batch_size, sequence_length), optional) — Float values of input raw speech waveform. Values can be obtained by loading a .flac or .wav audio file into an array of type list[float], a numpy.ndarray or a torch.Tensor, e.g. via the torchcodec library (pip install torchcodec) or the soundfile library (pip install soundfile). To prepare the array into input_values, the AutoProcessor should be used for padding and conversion into a tensor of type torch.FloatTensor. See MuseSparkProcessor.call() for details.
  • input_values_mask (torch.Tensor of shape (batch_size, num_samples), optional) — Which entries of input_values are real audio rather than padding, as produced by MuseSparkFeatureExtractor. Required when batching clips of different lengths.
  • mm_encoder_outputs (dict[str, ~modeling_outputs.BaseModelOutputWithPooling], optional) — Dict where keys are supported modalities and values are encoded outputs for that modality. Each encoded output is a tuple that consists of (pooler_output, optional: last_hidden_states, optional: hidden_states, optional: attentions) pooler_output of shape (batch_size, sequence_length, hidden_size), optional) is a sequence of multimmodal features of the encoder merged into text embeddings.

Returns

MuseSparkModelOutputWithPast or tuple(torch.FloatTensor)

A MuseSparkModelOutputWithPast 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 (MuseSparkConfig) and inputs.

The MuseSparkModel 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, sequence_length, hidden_size)) — Sequence of hidden-states at the output of the last layer of the model.

    If past_key_values is used only the last hidden-state of the sequences of shape (batch_size, 1, hidden_size) is output.

  • past_key_values (Cache, optional, returned when use_cache=True is passed or when config.use_cache=True) — It is a Cache instance. For more details, see our kv cache guide.

    Contains pre-computed hidden-states (key and values in the self-attention blocks and optionally if config.is_encoder_decoder=True in the cross-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.

  • image_hidden_states (torch.FloatTensor, optional) — A torch.FloatTensor of size (batch_size, num_images, sequence_length, hidden_size). image_hidden_states of the model produced by the vision encoder and after projecting the last hidden state.

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

    Router logits of the model, useful to compute the auxiliary loss for Mixture of Experts models.

Example:

>>> import os

>>> from transformers import AutoProcessor, DistributedConfig, MuseSparkModel

>>> distributed_config = DistributedConfig(
...     tp_size=int(os.environ["WORLD_SIZE"]),
...     ep_size=int(os.environ["WORLD_SIZE"]),
... )

>>> processor = AutoProcessor.from_pretrained("meta-models/Muse-Spark-Open-816B-A42B")
>>> model = MuseSparkModel.from_pretrained(
...     "meta-models/Muse-Spark-Open-816B-A42B",
...     dtype="auto",
...     distributed_config=distributed_config,
... )

>>> messages = [
...     {
...         "role": "user",
...         "content": [
...             {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/pipeline-cat-chonk.jpeg"},
...             {"type": "text", "text": "What is shown in this image?"},
...         ],
...     },
... ]

>>> inputs = processor.apply_chat_template(
...     messages,
...     reasoning_effort="medium",
...     add_generation_prompt=True,
...     tokenize=True,
...     return_dict=True,
...     return_tensors="pt",
... ).to(model.device).to(model.dtype)

>>> outputs = model(**inputs)
>>> last_hidden_state = outputs.last_hidden_state

MuseSparkForConditionalGeneration

class transformers.MuseSparkForConditionalGeneration

< >

( config )

Parameters

  • config (MuseSparkForConditionalGeneration) — 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 Muse Spark Model for token generation conditioned on other modalities (e.g. image-text-to-text generation).

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] = Nonepixel_values: typing.Optional[torch.FloatTensor] = Noneimage_grid_thw: typing.Optional[torch.LongTensor] = Nonepixel_values_videos: typing.Optional[torch.FloatTensor] = Nonevideo_grid_thw: typing.Optional[torch.LongTensor] = Noneinput_values: typing.Optional[torch.Tensor] = Noneinput_values_mask: typing.Optional[torch.Tensor] = Noneattention_mask: typing.Optional[torch.Tensor] = Noneposition_ids: typing.Optional[torch.LongTensor] = Nonepast_key_values: transformers.cache_utils.Cache | None = Noneinputs_embeds: typing.Optional[torch.FloatTensor] = Nonelabels: typing.Optional[torch.LongTensor] = Noneuse_cache: bool | None = Nonelogits_to_keep: typing.Union[int, torch.Tensor] = 0mm_encoder_outputs: dict[str, transformers.modeling_outputs.BaseModelOutputWithPooling] | None = None**kwargs: *<class 'transformers.utils.generic.TransformersKwargs'> ) → MuseSparkCausalLMOutputWithPast or tuple(torch.FloatTensor)

Parameters

  • 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. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.

    What are input IDs?

  • pixel_values (torch.FloatTensor of shape (batch_size, num_channels, image_size, image_size), optional) — The tensors corresponding to the input images. Pixel values can be obtained using MuseGlimmerImageProcessor. See MuseGlimmerImageProcessor.call() for details (MuseSparkProcessor uses MuseGlimmerImageProcessor for processing images).
  • image_grid_thw (torch.LongTensor of shape (num_images, 3), optional) — The temporal, height and width of feature shape of each image in LLM.
  • pixel_values_videos (torch.FloatTensor of shape (batch_size, num_frames, num_channels, frame_size, frame_size), optional) — The tensors corresponding to the input video. Pixel values for videos can be obtained using MuseGlimmerVideoProcessor. See MuseGlimmerVideoProcessor.call() for details (MuseSparkProcessor uses MuseGlimmerVideoProcessor for processing videos).
  • video_grid_thw (torch.LongTensor of shape (num_videos, 3), optional) — The temporal, height and width of feature shape of each video in LLM.
  • input_values (torch.Tensor of shape (batch_size, sequence_length), optional) — Float values of input raw speech waveform. Values can be obtained by loading a .flac or .wav audio file into an array of type list[float], a numpy.ndarray or a torch.Tensor, e.g. via the torchcodec library (pip install torchcodec) or the soundfile library (pip install soundfile). To prepare the array into input_values, the AutoProcessor should be used for padding and conversion into a tensor of type torch.FloatTensor. See MuseSparkProcessor.call() for details.
  • input_values_mask (torch.Tensor of shape (batch_size, num_samples), optional) — Which entries of input_values are real audio rather than padding, as produced by MuseSparkFeatureExtractor. Required when batching clips of different lengths.
  • 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?

  • 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_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 instance is allowed as input, see our kv cache guide. If no past_key_values are passed, 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).

  • 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].
  • 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).
  • mm_encoder_outputs (dict[str, ~modeling_outputs.BaseModelOutputWithPooling], optional) — Dict where keys are supported modalities and values are encoded outputs for that modality. Each encoded output is a tuple that consists of (pooler_output, optional: last_hidden_states, optional: hidden_states, optional: attentions) pooler_output of shape (batch_size, sequence_length, hidden_size), optional) is a sequence of multimmodal features of the encoder merged into text embeddings.

Returns

MuseSparkCausalLMOutputWithPast or tuple(torch.FloatTensor)

A MuseSparkCausalLMOutputWithPast 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 (MuseSparkConfig) and inputs.

The MuseSparkForConditionalGeneration 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.text_config.vocab_size)) — Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).

  • past_key_values (Cache, optional, returned when use_cache=True is passed or when config.use_cache=True) — It is a Cache instance. For more details, see our kv cache guide.

    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.

  • image_hidden_states (torch.FloatTensor, optional) — A torch.FloatTensor of size (batch_size, num_images, sequence_length, hidden_size). image_hidden_states of the model produced by the vision encoder after projecting last hidden state.

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

    Router logits of the model, useful to compute the auxiliary loss for Mixture of Experts models.

Example:

>>> import os

>>> from transformers import AutoProcessor, DistributedConfig, MuseSparkForConditionalGeneration

>>> distributed_config = DistributedConfig(
...     tp_size=int(os.environ["WORLD_SIZE"]),
...     ep_size=int(os.environ["WORLD_SIZE"]),
... )

>>> processor = AutoProcessor.from_pretrained("meta-models/Muse-Spark-Open-816B-A42B")
>>> model = MuseSparkForConditionalGeneration.from_pretrained(
...     "meta-models/Muse-Spark-Open-816B-A42B",
...     dtype="auto",
...     distributed_config=distributed_config,
... )

>>> messages = [
...     {
...         "role": "user",
...         "content": [
...             {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/pipeline-cat-chonk.jpeg"},
...             {"type": "text", "text": "What is shown in this image?"},
...         ],
...     },
... ]

>>> inputs = processor.apply_chat_template(
...     messages,
...     reasoning_effort="medium",
...     add_generation_prompt=True,
...     tokenize=True,
...     return_dict=True,
...     return_tensors="pt",
... ).to(model.device).to(model.dtype)
>>> input_len = inputs["input_ids"].shape[-1]

>>> outputs = model.generate(**inputs, max_new_tokens=512)

>>> response = processor.decode(outputs[0][input_len:], skip_special_tokens=False)
>>> print(response)
The image shows a **Pallas's cat — also known as a manul, *Otocolobus manul*** — walking through snow.
...
Update on GitHub

←MT5 MuseGlimmer→