load_nemotron_ssm

metrale_model_layers::weight_map::nemotron

Function load_nemotron_ssm 

Source 
pub fn load_nemotron_ssm(
    store: &WeightStore,
    _layer: usize,
    gpu: &dyn GpuBackend,
    layer_prefix: &str,
) -> Result<(NemotronSsmWeights, NemotronSsmQuant)>
Expand description

2026-09-25: Load a Nemotron-H Mamba-2 mixer ({layer_prefix}.mixer).

The format is read from in_proj alone. Unless it is NVFP4, in_proj and out_proj come back as QuantizedWeight::null(), and the returned NemotronSsmQuant tells the caller how to load them.