load_attention

metrale_model_layers::weight_map::loaders_fp8

Function load_attention 

Source 
pub fn load_attention(
    store: &WeightStore,
    layer_prefix: &str,
    gpu: &dyn GpuBackend,
    variant: Nvfp4Variant,
    qctx: QuantizeCtx,
    config: &ModelConfig,
) -> Result<AttentionWeights>
Expand description

2026-09-25: Load the {layer_prefix}.self_attn weights of a full-attention layer.