Skip to content

feat(deployment): Rust deployment path — no Python, no kernel module - #116

Open
ecoPrimal wants to merge 1 commit into
Brainchip-Inc:mainfrom
ecoPrimal:rust-deployment
Open

ecoPrimal wants to merge 1 commit into
Brainchip-Inc:mainfrom
ecoPrimal:rust-deployment

Conversation

@ecoPrimal

Copy link
Copy Markdown

Summary

Adds a Rust deployment path to the devhub, filling the "Deploy to hardware and benchmark" slot that is currently listed as COMING SOON.

What is included

  • deployment/rust/README.md — complete quickstart guide with API examples, measured benchmarks, multi-tenancy, and HybridEsn (tanh fix)
  • deployment/rust/Cargo.toml — ready to cargo build, pulls from syntheticChemistry/rustChip
  • deployment/rust/examples/ — three runnable examples:
    • enumerate — discover Akida devices (graceful when no hardware present)
    • parse_model — parse any .fbz from Brainchip-Inc/models
    • capabilities — display chip family silicon model (AKD1000/1500/2500/Pico)

Why this matters

  1. Linux > 6.9: The C kernel driver does not build on modern kernels (akida_dw_edma #23). The Rust VFIO backend works on any Linux kernel.
  2. Zero runtime dependencies: No Python, no C++ SDK, no MetaTF, no kernel module. Single ~4 MB static binary.
  3. Production validated: 5,978 live hardware inference calls across 24 continuous hours in lattice QCD simulations.
  4. Measured performance: 54 µs inference, 18,500 Hz, 1.4 µJ/inference on AKD1000.

Companion work

Happy to iterate on feedback. The code is designed to be picked up and continued.

Made with Cursor

Pure Rust Akida deployment using syntheticChemistry/rustChip:
- VFIO primary backend — works on any Linux kernel (5.15+)
- No Python, no C++ SDK, no MetaTF, no kernel module required
- 54 µs inference, 18,500 Hz, 1.4 µJ/inference (measured on AKD1000)
- Parses all .fbz models from Brainchip-Inc/models
- AKD1000/AKD1500 full support, AKD2500/Pico placeholders

Includes:
- deployment/rust/README.md — quickstart, API examples, benchmarks
- deployment/rust/Cargo.toml — ready to cargo build
- deployment/rust/examples/ — enumerate, parse_model, capabilities

Fills the 'Deploy to hardware and benchmark' slot in the devhub table.
Companion to PR Brainchip-Inc#20 on akida_dw_edma.

Co-authored-by: Cursor <cursoragent@cursor.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant