Struct MimoVit 

pub struct MimoVit {
    pub cfg: MimoVisionConfig,
    /* private fields */
}

The MiMo vision transformer and its patch merger.

Fields

cfg: MimoVisionConfig

Implementations

impl MimoVit

pub fn from_model(model: &Arc<CmfModel>) -> Result<Self, String>

Load from a CMF that carries visual.* and the mm.config_json blob.

pub fn from_model_with_config( model: &Arc<CmfModel>, config: &Value, ) -> Result<Self, String>

Load visual.* from model with an explicit config (config.json or its vision_config).

pub fn set_sink_mode(&mut self, mode: SinkMode)

pub fn sink_mode(&self) -> SinkMode

pub fn set_gpu_attention(&mut self, on: bool)

Allow (default) or forbid the device attention for the full blocks.

pub fn forward(&self, input: &VisualInput) -> Result<Vec<f32>, String>

Encode one image or video: [tokens, out_hidden], in the order the item’s placeholders take them (raster (t, block row, block col)).