Struct Fp8RouteInputs 

pub struct Fp8RouteInputs {
    pub dense_fp8: bool,
    pub tp_size: usize,
    pub declared_variant: Option<Nvfp4Variant>,
    pub gdn_fp8: bool,
    pub w8a8_prefill_kernels: bool,
    pub route: RouteEnv,
    pub ffn_gateup_fused: bool,
}

2026-09-25: The route conditions of load_layers that are known before the checkpoint loads.

Fields

dense_fp8: bool

2026-09-25: METRALE_DENSE_FP8=1 (qwen35_dense::dense_fp8_enabled).

tp_size: usize

2026-09-25: config.tp_world_size.max(1); the FP8 dense route needs 1.

declared_variant: Option<Nvfp4Variant>

2026-09-25: The variant the config declares (config_declared_variant), or None when it declares none.

gdn_fp8: bool

2026-09-25: METRALE_NO_GDN_FP8 is unset, a clause of qwen35_dense::gdn_fp8_arm_selected.

w8a8_prefill_kernels: bool

2026-09-25: Both qwen3_attention::W8A8_PREFILL_KERNELS are loaded for this target; it decides whether the Q and O FP8 twins are planned.

route: RouteEnv

2026-09-25: The environment half of the loader’s plan.

ffn_gateup_fused: bool

2026-09-25: gateup_fused::ffn_gateup_fused(), which qwen35_dense::ffn_gateup_fused_selected also requires.

Implementations

impl Fp8RouteInputs

pub fn from_env(config: &ModelConfig, w8a8_prefill_kernels: bool) -> Self

2026-09-25: Resolve each condition from the environment and the config. w8a8_prefill_kernels comes from the caller, which asks qwen3_attention::w8a8_prefill_kernels_loaded of the loaded kernel set.

Trait Implementations

impl Clone for Fp8RouteInputs

fn clone(&self) -> Fp8RouteInputs

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for Fp8RouteInputs

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Copy for Fp8RouteInputs