bridge.models.muse_spark_open.muse_spark_open_config

Serializable configuration for the Muse Spark Open HybridModel integration.

Module Contents

Classes

MuseSparkTransformerConfig	Additional residual-gate configuration for the sparse decoder.
MuseSparkModelConfig	HF perception modules and an MCore Hybrid language model.

API

class bridge.models.muse_spark_open.muse_spark_open_config.MuseSparkTransformerConfig

Bases: megatron.bridge.models.muse_glimmer.muse_glimmer_config.MuseGlimmerTransformerConfig

Additional residual-gate configuration for the sparse decoder.

residual_gate_temperature: float

0.3

class bridge.models.muse_spark_open.muse_spark_open_config.MuseSparkModelConfig

Bases: megatron.bridge.models.common.ModelConfigOverrideMixin, megatron.bridge.models.hybrid.hybrid_builder.HybridModelConfig

HF perception modules and an MCore Hybrid language model.

builder: ClassVar[str]

‘megatron.bridge.models.muse_spark_open.MuseSparkModelBuilder’

transformer_config_class: ClassVar[type]

None

hf_config: dict[str, Any]

‘field(…)’

hf_model_text_only: bool

False

freeze_language_model: bool

False

freeze_vision_model: bool

False

freeze_audio_model: bool

False

freeze_vision_projection: bool

False

get_builder_cls

None

property special_token_ids: dict[str, int]

Return the media placeholder IDs for collators.

finalize() → None

Validate the first supported execution path before model allocation.

as_dict() → dict[str, Any]

Serialize callable activations using Bridge’s existing registry.

classmethod from_dict(

data: dict[str, Any],

) → bridge.models.muse_spark_open.muse_spark_open_config.MuseSparkModelConfig

Restore a serialized config without importing the Transformers model code.