Enum MoshiCheckpoint 

pub enum MoshiCheckpoint {
    Bf16Safetensors,
    Q8Gguf,
    Q4MlxSafetensors,
    Q8MlxSafetensors,
    MlxBf16Safetensors,
}

Published Moshi LM weight preset (quantization + on-disk layout).

Variants

Bf16Safetensors

kyutai/moshiko-candle-bf16 — model.safetensors (~14 GB).

Q8Gguf

kyutai/moshiko-candle-q8 — model.q8.gguf (~8 GB).

Q4MlxSafetensors

kyutai/moshiko-mlx-q4 — model.q4.safetensors (~4 GB).

Q8MlxSafetensors

kyutai/moshiko-mlx-q8 — model.q8.safetensors (~7 GB).

MlxBf16Safetensors

kyutai/moshiko-mlx-bf16 — model.safetensors (~14 GB, MLX key layout).

Implementations

impl MoshiCheckpoint

pub const SHARED_FILES: &'static [&'static str]

Shared tokenizer + mimi sidecar files (same across Kyutai repos).

pub fn hf_repo(self, voice: MoshiVoice) -> &'static str

HuggingFace model repo for this preset and voice family.

pub fn default_cache_dir(self, voice: MoshiVoice) -> PathBuf

Default .cache/… directory for a voice + preset pair (unless RLX_MOSHI_DIR overrides).

pub fn lm_filename(self) -> &'static str

Primary LM weights filename inside a model directory.

pub fn is_gguf(self) -> bool

True when weights are Kyutai Q8 GGUF (model.q8.gguf).

pub fn is_mlx(self) -> bool

MLX key layout (Q4/Q8/bf16 safetensors).

pub fn is_mlx_quantized(self) -> bool

Affine-quantized MLX weights (U32 packed + scales/biases).

pub fn mlx_quant_params(self) -> (u32, usize)

Bits per weight and group size for affine MLX quants; (0, 0) for full-precision MLX bf16.

pub fn gpu_loadable(self) -> bool

Whether the Candle moshi GPU backend can load this preset.

pub fn candle_compatible(self) -> bool

Candle-native tensor names (bf16 safetensors or Q8 GGUF).

pub fn size_class_gb(self) -> f32

Approximate published size class for docs / fetch hints.

pub fn parse(name: &str) -> Option<MoshiCheckpoint>

Parse CLI / env preset names (bf16, q8, q4, q8-mlx, mlx-bf16, …).

pub fn from_env_or_default() -> MoshiCheckpoint

Read RLX_MOSHI_CHECKPOINT, defaulting to Candle bf16 safetensors.

pub fn lm_weights_path(self, model_dir: &Path) -> PathBuf

Absolute path to LM weights inside model_dir.

pub fn tokenizer_path(self, model_dir: &Path) -> PathBuf

SentencePiece model path (shared across Kyutai repos).

Trait Implementations

impl Clone for MoshiCheckpoint

fn clone(&self) -> MoshiCheckpoint

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for MoshiCheckpoint

impl Debug for MoshiCheckpoint

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

Formats the value using the given formatter.

impl Eq for MoshiCheckpoint

impl Hash for MoshiCheckpoint

fn hash<__H>(&self, state: &mut __H)

where __H: Hasher,

Feeds this value into the given Hasher.

fn hash_slice<H>(data: &[Self], state: &mut H)

where H: Hasher, Self: Sized,

Feeds a slice of this type into the given Hasher.

impl PartialEq for MoshiCheckpoint

fn eq(&self, other: &MoshiCheckpoint) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl StructuralPartialEq for MoshiCheckpoint