turboquant-rs 0.2.0

TurboQuant KV-Cache Quantization — 3-bit compression with zero accuracy loss (Zandieh et al., ICLR 2026)
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31

//! TurboQuant -- KV-Cache quantization with zero accuracy loss.
//!
//! Implements Google's TurboQuant algorithm (Zandieh et al., ICLR 2026)
//! for compressing LLM key-value caches to 3-4 bits per value.

pub mod attention;
pub mod codebook;
pub mod error;
pub mod math;
pub mod packed;
pub mod qjl;
pub mod quantize;
pub mod rotation;

#[cfg(test)]
mod test_utils;

pub use attention::QuantizedKVCache;
pub use error::{Result, TurboQuantError};
pub use packed::PackedBlock;
pub use packed::TurboQuantConfig;
pub use qjl::{
    estimate_inner_product, estimate_inner_product_single, estimate_inner_product_with_codebook,
    precompute_query_projections, quantize_with_qjl, quantize_with_qjl_resources,
    EstimationContext, QjlBatchResources, QjlBlock,
};
pub use quantize::{
    dequantize_into_with_codebook, dequantize_vec, dequantize_vec_with_codebook, quantize_vec,
    quantize_vec_with_codebook,
};
pub use rotation::RotationOrder;