Home

nir-rs

Pure-Rust implementation of the Neuromorphic Intermediate Representation (NIR).

Version: 0.3.0 | Edition: 2024 | License: MIT OR Apache-2.0

NIR is to SNNs what ONNX is to conventional nets: a framework-agnostic graph format. This crate provides a typed in-memory graph, wire-accurate node types, and opt-in HDF5 .nir read/write interoperable with the Python reference.

Status

Milestone	Focus	Status
v0.1	License, skeleton, CI	Done
v0.2	Typed graph, nodes, errors	Done
v0.3	HDF5 I/O via hdf5-metno, fixtures	Current
v0.4	Serde/debug DX, examples	Planned
v0.5	Consumers (silicon-bridge, axon-encoder, …)	Planned

Start Here

Page	Description
Getting Started	Install, graph build, HDF5 feature
Project Structure	Modules and features
Architecture	Ownership and non-goals
NirGraph	In-memory graph API
Nodes	Closed NirNode enum / wire types
Tensor and Metadata	Dense tensors, dtype
File I/O	io::read / write, HDF5 layout
Wire Format	Keys, version, type strings
Errors	NirError
Testing and Fixtures	Python-vendored .nir files
CI and Quality	Actions, review bar
Ecosystem	Upstream and Limen stack
Glossary	Terms

Quick Commands

cargo test                 # graph only — no libhdf5
cargo test --all-features  # + HDF5 I/O
cargo clippy --all-targets --all-features -- -D warnings
cargo fmt --check

License

Dual-licensed under MIT or Apache-2.0. Vendored test fixtures under tests/fixtures/ are BSD-3 from neuromorphs/NIR.

---

Last updated: July 29, 2026 Updated by: Grok Build: Grok 4.5 Package tip reference: df72fc9 (main, v0.3 / PR #20)