ring-db 0.1.2

A Rust library for ring queries in high-dimensional vector spaces.
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
32
33
34
35
36

use crate::error::Result;

pub mod cpu;

pub use cpu::CpuBackend;

/// Trait implemented by all compute backends.
///
/// The separation between upload and query reflects the core performance
/// design: the dataset is uploaded once, then many queries run against
/// the resident data without re-uploading.
pub trait RingComputeBackend: Send + Sync {
    /// Human-readable backend name (e.g. `"cpu"`).
    fn name(&self) -> &'static str;

    /// Upload a float32 dataset to the backend.
    ///
    /// - `dims`: number of dimensions per vector.
    /// - `vectors`: flat row-major buffer, length `n * dims`.
    /// - `norms_sq`: precomputed squared L2 norm per vector, length `n`.
    ///
    /// Ownership is transferred: the backend decides whether to keep or
    /// consume the data (e.g. a GPU backend would copy to VRAM and drop).
    fn upload_f32_dataset(
        &mut self,
        dims: usize,
        vectors: Vec<f32>,
        norms_sq: Vec<f32>,
    ) -> Result<()>;

    /// Execute a float32 ring search.
    ///
    /// Returns IDs of all vectors with Euclidean distance to `query` in
    /// `[d - lambda, d + lambda]`.
    fn ring_query_f32(&self, dims: usize, query: &[f32], d: f32, lambda: f32) -> Result<Vec<u32>>;
}