Struct BitNetConfig 

pub struct BitNetConfig {
    pub group_size: usize,
    pub activation_bits: u8,
    pub per_token_activation: bool,
    pub use_rms_norm: bool,
    pub eps: f32,
    pub enable_ste: bool,
}

Configuration for BitNet b1.58 quantization.

BitNet uses:

• Ternary weights: {-1, 0, +1} via AbsMean quantization
• INT8 activations: Per-token AbsMax scaling

Reference

“The Era of 1-bit LLMs: All Large Language Models are in 1.58 Bits” https://arxiv.org/abs/2402.17764

Fields

group_size: usize

Group size for weight quantization. Weights are quantized in groups, each with its own scale. Typical values: 64, 128, 256.

activation_bits: u8

Number of bits for activation quantization. BitNet b1.58 uses 8 bits (INT8).

per_token_activation: bool

Whether to use per-token activation scaling. If true, each token gets its own scale factor. If false, uses per-tensor scaling.

use_rms_norm: bool

Whether to apply RMS normalization before quantization.

eps: f32

Epsilon for numerical stability in normalization.

enable_ste: bool

Whether to enable Straight-Through Estimator for training.

Implementations

impl BitNetConfig

pub fn new() -> Self

Create a new configuration with default values.

pub fn inference() -> Self

Create configuration optimized for inference.

Disables training-specific features like STE.

pub fn training() -> Self

Create configuration for training.

Enables STE for gradient estimation through quantization.

pub const fn with_group_size(self, group_size: usize) -> Self

Set the group size for weight quantization.

pub const fn with_activation_bits(self, bits: u8) -> Self

Set the activation bit width.

pub const fn with_per_token_activation(self, enabled: bool) -> Self

Enable or disable per-token activation scaling.

pub const fn with_rms_norm(self, enabled: bool) -> Self

Enable or disable RMS normalization.

pub const fn with_ste(self, enabled: bool) -> Self

Enable or disable Straight-Through Estimator.

pub fn validate(&self) -> Result<()>

Validate the configuration.

Errors

Returns error if configuration is invalid.

Trait Implementations

impl Clone for BitNetConfig

fn clone(&self) -> BitNetConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for BitNetConfig

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

Formats the value using the given formatter.

impl Default for BitNetConfig

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for BitNetConfig

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for BitNetConfig

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.