Skip to main content

Crate neurolib

Crate neurolib 

Source
Expand description

§NeuroLib

Single entry point to the workspace: one crate, one Cargo.toml input, everything else is included through features - in the same way that crate burn collects its subcrates.

[dependencies]
neurolib = { version = "0.1", features = ["snn", "wgpu"] }
ModuleCrateFeature
snnneo-snnsnn
nnneo-nnnn
quantneo-quantizationquant
synkneo-synkkernels
storeneo-storestore
coreneo-corecore
ioneo-ioio (zarr, neuro)
viewneo-viewview
synapseneo-synapsesynapse

Groups: ml (default), signal, full. GPU: wgpu, cuda, rocm, plus fusion. neo-gui is an application, not a library, so it’s not here.

Kernels are reachable from the spiking module, no need to name neo-synk:

ⓘ
use burn::tensor::{Device, Tensor};
use neurolib::snn::leaky_scan_forward;

let device = Device::wgpu(Default::default());
let u = leaky_scan_forward(input, mem0, beta); // wgpu/CUDA/ROCm: fused CubeCL kernel

burn is not re-exported: declare it separately in the project (burn = "0.22", as in workspace).

Re-exports§

pub use neo_snn as snn;
pub use neo_nn as nn;
pub use neo_quantization as quant;
pub use neo_synk as synk;
pub use neo_store as store;
pub use neo_core as core;
pub use neo_io as io;
pub use neo_view as view;
pub use neo_synapse as synapse;

Modules§

prelude
Everyday names: currently prelude neo-snn. The remaining crates are imported via own modules (neurolib::nn::...) so that glob import does not produce name conflicts.