Expand description
This crate provides high-level bindings to Faiss, the vector similarity search engine.
§Preparing
This crate requires Faiss and the C API to be built beforehand by the developer. Please follow the instructions here, and build the dynamic library with the C API (instructions here)
This will result in the dynamic library faiss_c (“libfaiss_c.so” in Linux),
which needs to be installed in a place where your system will pick up. In
Linux, try somewhere in the LD_LIBRARY_PATH environment variable, such as
“/usr/lib”, or try adding a new path to this variable.
§GPU support
In order to have GPU support, the gpufaiss_c library from the main project
needs to be built instead. Then, enable the gpu Cargo feature for this crate.
[dependencies]
"faiss" = {version = "0.12.0", features = ["gpu"]}§Examples
The Index trait is one of the center-pieces of this library. Index
implementations can be requested using the index_factory function:
use faiss::{Index, index_factory, MetricType};
let mut index = index_factory(8, "Flat", MetricType::L2)?;
index.add(my_data)?;
let result = index.search(my_query, 5)?;
for (i, (l, d)) in result.labels.iter()
.zip(result.distances.iter())
.enumerate()
{
println!("#{}: {} (D={})", i + 1, *l, *d);
}With GPU support, create a StandardGpuResources and use the
into_gpu and into_cpu methods to move an index to and from the GPU.
use faiss::{GpuResources, StandardGpuResources, Index, index_factory, MetricType};
let index = index_factory(8, "Flat", MetricType::L2)?;
let gpu_res = StandardGpuResources::new()?;
let index = index.into_gpu(&gpu_res, 0)?;Unless otherwise indicated, vectors are added and retrieved from the
library under the form of contiguous column-first slices of f32 elements.
Details from the official Faiss APIs still apply. Please visit the Faiss wiki for additional guidance.
Re-exports§
pub use index::flat::FlatIndex;pub use index::id_map::IdMap;pub use index::io::read_index;pub use index::io::write_index;pub use index::lsh::LshIndex;pub use index::index_factory;pub use index::ConcurrentIndex;pub use index::Idx;pub use index::Index;pub use metric::MetricType;pub use gpu::GpuResources;pub use gpu::StandardGpuResources;pub use index::gpu::GpuIndexImpl;