Enum ModelVariant 

pub enum ModelVariant {
    Bonsai8B,
    Bonsai4B,
    Bonsai1_7B,
    TernaryBonsai8B,
    TernaryBonsai4B,
    TernaryBonsai1_7B,
    FP8Bonsai8B,
    FP8Bonsai4B,
    FP8Bonsai1_7B,
    Custom,
}

Known Bonsai model variants.

Variants

Bonsai8B

Bonsai-8B (Qwen3-8B architecture): 36 layers, hidden=4096

Bonsai4B

Bonsai-4B: 24 layers, hidden=2560

Bonsai1_7B

Bonsai-1.7B: 16 layers, hidden=1536

TernaryBonsai8B

Ternary-Bonsai-8B: same Qwen3-8B architecture, {-1,0,+1} weights (TQ2_0_g128).

TernaryBonsai4B

Ternary-Bonsai-4B: same Qwen3-4B architecture, {-1,0,+1} weights (TQ2_0_g128).

TernaryBonsai1_7B

Ternary-Bonsai-1.7B: same Qwen3-1.7B architecture, {-1,0,+1} weights (TQ2_0_g128).

FP8Bonsai8B

FP8-Bonsai-8B: same Qwen3-8B architecture, FP8 weights (F8_E4M3 or F8_E5M2).

FP8Bonsai4B

FP8-Bonsai-4B: same Qwen3-4B architecture, FP8 weights.

FP8Bonsai1_7B

FP8-Bonsai-1.7B: same Qwen3-1.7B architecture, FP8 weights.

Custom

Custom or unrecognized architecture

Implementations

impl ModelVariant

pub fn from_config(config: &Qwen3Config) -> Self

Auto-detect variant from model configuration.

Matches on the combination of num_layers and hidden_size to identify known architectures.

pub fn from_config_and_sample_tensor_type( config: &Qwen3Config, sample_tensor_type: GgufTensorType, ) -> Self

Detect model variant from config + sample tensor type (for ternary vs 1-bit disambiguation).

Architecture match is identical to from_config, but if sample_tensor_type.is_ternary(), the result is upgraded to the ternary sibling variant.

pub fn default_config(&self) -> Qwen3Config

Get the default configuration for this variant.

Returns the standard configuration for known variants. For Custom, returns the 8B configuration as a fallback.

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

Human-readable display name for this variant.

pub fn param_count(&self) -> u64

Approximate parameter count for this variant.

Computed as: embedding + attention + ffn + norms + output head. For 1-bit models, each “parameter” is 1 bit + per-group scale. Ternary variants share the same architecture (and thus the same parameter count) as their 1-bit siblings; only the storage format differs.

pub fn expected_model_size_bytes(&self) -> u64

Expected model file size in bytes for the quantized GGUF file.

For 1-bit variants: ~1 bit per param + scale factors + FP16 embeddings. For ternary variants: TQ2_0_g128 uses 34 bytes per 128 weights ≈ 0.266 bytes/param. Embeddings and norms are typically stored in FP16 or FP32.

pub fn known_variants() -> &'static [ModelVariant]

Return all known (non-Custom) variants.

pub fn is_known(&self) -> bool

Whether this variant is a known (non-custom) architecture.

Trait Implementations

impl Clone for ModelVariant

fn clone(&self) -> ModelVariant

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ModelVariant

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

Formats the value using the given formatter.

impl Display for ModelVariant

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

Formats the value using the given formatter.

impl Hash for ModelVariant

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

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 ModelVariant

fn eq(&self, other: &ModelVariant) -> 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 Copy for ModelVariant

impl Eq for ModelVariant

impl StructuralPartialEq for ModelVariant