llama4_maverick_17b_128e¶
- torchtune.models.llama4.llama4_maverick_17b_128e(decoder_trainable: bool = True, encoder_trainable: bool = False, fusion_trainable: bool = True, image_size: int = 336) EarlyFusionModel[source]¶
Builder for creating an instance of the Llama4 Maverick 17Bx128E model
- Parameters:
decoder_trainable (bool) – Whether to make decoder params trainable. Default is True.
encoder_trainable (bool) – Whether to make encoder params trainable. Default is False.
fusion_trainable (bool) – Whether to make fusion params trainable. Default is True.
image_size (int) – Base image size that images will be tiled and resized to. Default is 336.
- Returns:
Instantiation of a 17Bx128E Llama4 MoE model with encoders.
- Return type:
EarlyFusionModel
