Aero v1.0.0

The First Complete AI-Native Systems Language

Rust-level safety • C-level speed • Python-level ergonomics for AI and numerical computing.

Aero solves the two-language problem by providing a single, memory-safe systems language that delivers production-grade deep learning performance without Python or garbage collection overhead.

🚀 Live Ecosystem

• Interactive WebAssembly Playground – Compile and run Aero code directly in your browser
• Benchmarking Dashboard – Live performance telemetry
• Full Documentation – "NumPy to AeroNum in 10 minutes" and API reference

⚡ Performance Highlights

• ≥1.4× faster than PyTorch on end-to-end GPT-2 mini-transformer training (1047 vs 748 tokens/sec)
• ≥5.3× GPU acceleration on 4096×4096 matrix multiplication with automatic .to("cuda") dispatch
• Near-linear multi-GPU scaling via native NCCL/MPI distributed training (up to 8 GPUs)
• Zero garbage collection pauses – full ownership model with compile-time safety

🧠 Why Aero Wins

1. Zero-Cost Abstractions – Neural networks compile to native code with no runtime overhead
2. Memory Safety by Construction – Ownership and borrowing eliminate use-after-free and data races
3. Familiar Ergonomics – Sequential, Dense, to("cuda"), save()/load() feel like PyTorch
4. Distributed by Default – Native DataParallel and ModelParallel with zero-copy NCCL communication

📦 Quick Start

git clone https://github.com/RobVanProd/Aero.git
cd Aero
cargo build --release
export PATH="$PWD/target/release:$PATH"

# Initialize a new project scaffold
aero init my_app
cd my_app

# Compile + run
aero run src/main.aero

# ROCm-targeted compile path (RX 7800 XT / gfx1101)
aero run --target rocm --gpu gfx1101 src/main.aero

# Backend alias form (equivalent to --target)
aero run --backend rocm --gpu gfx1101 src/main.aero

# Auto-detect local GPU backend (ROCm/CUDA/CPU fallback)
aero run --target gpu src/main.aero

# Type-check only (no codegen)
aero check src/main.aero

# Generate Markdown API docs from source
aero doc src/main.aero -o main.md

# Profile compilation pipeline and export trace JSON
aero profile src/main.aero -o trace.json

# Apply graph compilation with executable fusion (CPU/CUDA/ROCm)
aero graph-opt main.ll -o main.opt.ll --backend rocm --gpu gfx1101

# Apply hardware-calibrated quantization lowering (INT8/FP8)
aero quantize main.opt.ll -o main.int8.ll --mode int8 --backend rocm --gpu gfx1101 --calibration calib.json

# Run cross-framework GGUF benchmark harness (Aero vs llama.cpp vs PyTorch)
python benchmarks/gguf/gguf_compare.py --config benchmarks/gguf/config.rx7800xt.example.json

# Registry search (offline index or live transport)
aero registry search vision --live --registry https://registry.aero/api/v1

# Run formal conformance + mechanized checks
aero conformance -o conformance_report.json

# Language server for editor integration (stdio)
aero lsp

Try the flagship example directly in the Interactive Playground:

use aeronum::Array;
use aeronn::{Transformer, Sequential};

fn main() {
    let mut model = Transformer::new(layers: 6, dim: 384, heads: 6);
    model.to("distributed", 4);  // Scale across 4 GPUs
    // Train at native speed...
}

🛠️ Compiler Features (v1.0.0)

Category	Features
Type System	Static typing, generics, trait bounds, where clauses
Memory	Ownership, move semantics, shared & mutable references, borrow checker
Data Types	Structs, enums, arrays, tuples, strings, pattern matching
Control Flow	Functions, if/else, while/for loops, break/continue, closures
Modules	mod/use imports, pub visibility, multi-file projects
Codegen	LLVM IR backend with optimization passes
CLI	aero build, aero run, aero check, aero test, aero fmt, aero doc, aero profile, aero graph-opt, aero quantize, aero registry, aero conformance, aero init, aero lsp
LSP	Syntax diagnostics, completion, hover, go-to-definition, document symbols
Docs & Profiling	Markdown API generation (aero doc), compilation stage timing + trace export (aero profile)
Phase 8 Runtime Slice	Hardware-calibrated INT8/FP8 lowering (CPU/CUDA/ROCm), executable fused-kernel backend generation, live registry.aero transport/auth/trust model, formal conformance + mechanized checks
Diagnostics	Colored errors, source snippets, "did you mean?" suggestions

Formal spec: docs/language/aero_formal_language_specification.md

Looking Ahead

• GGUF-native model loader and runtime benchmarks on CUDA/ROCm
• Expanded optimizer and fused-kernel library coverage
• Additional formal semantics proofs beyond deterministic conformance checks

License

MIT © RobVanProd and contributors. See LICENSE for details.