Skip to content

Add NEURAX - Analytical compiler for neural network architectures - #31

Open
rustnew wants to merge 1 commit into
vaaaaanquish:mainfrom
rustnew:add-neurax
Open

Add NEURAX - Analytical compiler for neural network architectures#31
rustnew wants to merge 1 commit into
vaaaaanquish:mainfrom
rustnew:add-neurax

Conversation

@rustnew

@rustnew rustnew commented Aug 7, 2026

Copy link
Copy Markdown

NEURAX - Analytical Compiler for Neural Architectures

Repository: https://github.com/rustnew/NEURAX

Features

  • 10-pass analytical IR pipeline (Rust)
  • MLIR backend with 13 custom dialects (LLVM 18)
  • 11 architecture families: Transformer, MoE, CNN, SSM, Diffusion, GNN, GAN, RL, SNN, RNN, Experimental
  • Predicts FLOPs, VRAM, training cost, energy, carbon emissions
  • <50ms analysis, no GPU required
  • React 18 visual canvas + CLI + TUI
  • 680+ configurable blocks, 88 reference templates

Why Rust?

  • Core engine (neurax-core, neurax-ir, neurax-parser) written in Rust
  • Safety + performance critical for analytical compilation
  • MLIR bindings via Rust
  • 99%+ accuracy on predictions

Architecture

  • neurax-core - Pipeline orchestrator
  • neurax-ir - 10-dialect analytical IR
  • neurax-formulas - Analytical formulas
  • neurax-hardware-db - GPU/CPU specs
  • neurax-mlir - MLIR backend (13 dialects)
  • neurax-cli - CLI tool

Links

MIT licensed. Contributions welcome!

- Analytical compiler for neural network architectures
- Predicts training cost, memory, performance before training
- <50ms analysis, zero GPU, 99%+ accuracy
- MLIR/LLVM 18 backend, 13 custom dialects
- MIT licensed, open source
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