Struct KvLayerVerdict 

pub struct KvLayerVerdict {
    pub layer: usize,
    pub stream: &'static str,
    pub n_vectors: usize,
    pub recon_int8: f64,
    pub int8_bits: f64,
    pub recon_dict: f64,
    pub dict_code_bits: f64,
    pub matched_bits: u32,
    pub recon_uniform_matched: f64,
    pub go: bool,
}

Per-(layer, stream) rate-distortion comparison. int8 (the W5a incumbent, 8 bits/elem) is a strong, accurate baseline; a sparse dictionary trades accuracy for a much smaller footprint. So the decision isn’t “does the dictionary beat int8’s accuracy” (it won’t — int8 has ~4× the bits) but “at the dictionary’s OWN low bit rate, does the learned basis beat NAIVE uniform quantization?” — i.e. is the learned codebook worth more than just quantizing more coarsely.

Fields

layer: usize

stream: &'static str

“K” or “V”.

n_vectors: usize

recon_int8: f64

int8 incumbent (8-bit) reconstruction error and footprint — context, not the gate.

int8_bits: f64

recon_dict: f64

k-sparse dictionary reconstruction error and asymptotic code footprint (bits/vec, dictionary amortized away as at deployment scale).

dict_code_bits: f64

matched_bits: u32

Naive uniform quantization at the dictionary’s matched bit rate — the head-to-head baseline.

recon_uniform_matched: f64

go: bool

GO here: the learned dictionary reconstructs at least as well as uniform quantization AT THE SAME bit rate (the learned basis earns its keep).

Trait Implementations

impl Clone for KvLayerVerdict

fn clone(&self) -> KvLayerVerdict

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for KvLayerVerdict

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

Formats the value using the given formatter.