seli-vector-db
| Crates.io | seli-vector-db |
| lib.rs | seli-vector-db |
| version | 0.1.2 |
| created_at | 2025-12-01 18:02:49.693653+00 |
| updated_at | 2025-12-03 07:40:51.964431+00 |
| description | A simple vector database in Rust using an IVF index for approximate nearest neighbor search |
| homepage | |
| repository | |
| max_upload_size | |
| id | 1960228 |
| size | 38,823 |
Sergey (Selithrarion)
documentation
README
$env:RUSTFLAGS='-C target-cpu=native'; cargo bench
1000 vectors, k 10, brute force - 123 µs
1000 vectors, k 10, brute force + normalize - 95 µs
1000 vectors, k 10, ivf_1k_nprobe_5 - 30 µs
100.000 vectors, k 10, brute force + normalize - 18 ms
100.000 vectors, k 10, ivf_1k_nprobe_5 - 225 µs
10.000 vectors, k 10, brute force + normalize - 1.5 ms
10.000 vectors, k 10, ivf_1k_nprobe_5 - 86 µs
10.000 vectors, k 10, brute force + normalize - 1.5 ms
10.000 vectors, k 10, ivf_1k_nprobe_5 + inline always for dot product - 80 µs
10.000 vectors, k 10, brute force + normalize - 1.5 ms
10.000 vectors, k 10, ivf_1k_nprobe_5 + vec for index - 83 µs
10.000 vectors, k 10, brute force + simd rustflag - 1.5 ms
10.000 vectors, k 10, ivf_1k_nprobe_5 + simd rustflag 73 µs
10.000 vectors, k 10, brute force + wide for simd (dot product optimized) - 405 µs
10.000 vectors, k 10, ivf_1k_nprobe_5 + wide for simd (dot product optimized) - 20-25 µs
normalize simd optimization nothing changed, became kinda worse
flat memory layout (Vec