load_dense_ffn

metrale_model_layers::weight_map::fp8_lut

Function load_dense_ffn 

Source 
pub fn load_dense_ffn(
    store: &WeightStore,
    prefix: &str,
    gpu: &dyn GpuBackend,
    variant: Nvfp4Variant,
    absmax_k: KernelHandle,
    quantize_k: KernelHandle,
    stream: u64,
    config: &ModelConfig,
) -> Result<DenseFfnWeights>
Expand description

2026-09-25: Load a dense (non-MoE) FFN’s gate/up/down projections as NVFP4, with transposed copies of each.