Struct BackboneLoadOptions 

pub struct BackboneLoadOptions {
    pub metal_prefill: MetalGgufPrefillMode,
    pub use_fast_kv: Option<bool>,
    pub memory_efficient: bool,
}

Options when constructing crate::backbone::BackboneModel.

Fields

metal_prefill: MetalGgufPrefillMode

Metal GGUF prefill strategy for the KV generator path.

use_fast_kv: Option<bool>

Force incremental KV decode vs packed runner. None uses env (ORPHEUS_METAL_KV, ORPHEUS_PACKED_LM, …).

memory_efficient: bool

When true (default for Self::synthesis), use dynamic decode and skip the bucket compile ladder unless ORPHEUS_BUCKET_DECODE=1.

Implementations

impl BackboneLoadOptions

pub fn synthesis() -> BackboneLoadOptions

Low-RAM TTS: CPU F32 GGUF prefill + CPU decode on Metal hosts (reference path).

pub fn gpu(device: Device) -> BackboneLoadOptions

GPU TTS on Metal / CUDA / ROCm: on-device prefill + decode (mmap GGUF). wgpu / Vulkan use CPU GGUF prefill + decode (parity) with SNAC logit mask.

pub fn for_device(device: Device) -> BackboneLoadOptions

Pick GPU path when an accelerator is available and low-memory mode is off.

pub fn reference_parity() -> BackboneLoadOptions

Reference parity / throughput: bucket decode allowed when RAM budget permits.

pub fn fast_prefill() -> BackboneLoadOptions

Experimental: packed Metal KV + CPU F32 logits.

pub fn with_memory_efficient(self, enabled: bool) -> BackboneLoadOptions

pub fn with_metal_prefill( self, mode: MetalGgufPrefillMode, ) -> BackboneLoadOptions

pub fn with_fast_kv(self, enabled: bool) -> BackboneLoadOptions

Trait Implementations

impl Clone for BackboneLoadOptions

fn clone(&self) -> BackboneLoadOptions

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for BackboneLoadOptions

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

Formats the value using the given formatter.

impl Default for BackboneLoadOptions

fn default() -> BackboneLoadOptions

Returns the “default value” for a type.