SciRS2 0.4.0 Released — Pure Rust SciPy Replacement Now at 2.91 Million SLoC

The scientific computing and AI foundation of the COOLJAPAN ecosystem just took a massive leap forward.

Today we released SciRS2 0.4.0 — a complete, production-grade Pure Rust framework that delivers SciPy-compatible APIs for scientific computing, data analysis, and machine learning.

No C. No Fortran. No NumPy system dependencies.
No build hell. No version conflicts.
Just clean, memory-safe, blazing-fast code that compiles to a single static binary (or WASM) and runs everywhere — from laptops to browsers to edge devices to cloud clusters.

Why SciRS2 0.4.0 is a game changer

For decades, scientific computing and AI meant depending on NumPy/SciPy (built on C/Fortran) or heavy Python ecosystems.

These tools are powerful but suffer from:

• Complex system dependencies and installation nightmares
• Memory unsafety and data races in native code
• Platform-specific binaries and vendor lock-in
• Poor performance in WASM, embedded, or no_std environments
• Slow Python loops for heavy numerical and AI work

SciRS2 0.4.0 ends all of that.

It delivers 10–100× faster performance while being 100% memory-safe and portable.
Notable results (Apple M3 / x86_64):

• Element-wise operations: 14.17× faster
• Signal convolution: 25× faster
• Bootstrap sampling: 30× faster
• FFT (fractional): 24.9× faster
• Matrix multiply (1000×1000): 1.25× faster
• SVD (500×500): 1.15× faster

Technical Deep Dive: How We Built a Production-Grade SciPy Replacement in Pure Rust

The architecture is modular (29 workspace crates) and built directly on the COOLJAPAN stack with full Pure Rust migration:

1. Core Layer (scirs2-core)
   NUMA-aware scheduler, lock-free data structures, GPU backends (CUDA/Metal/WebGPU), cache-oblivious algorithms.

2. Scientific Computing
   Full OxiBLAS integration, OxiFFT (cache-oblivious + adaptive sparse), GPU PDE solvers, GPU SpMV, stats, optimization, interpolation (Deep Kriging, GP Surrogate), signal processing, special functions.

3. AI/ML Layer (major 0.4.0 upgrades)
   Flash Attention 2, QAT, LoRA/DoRA/GPTQ, ONNX export, Temporal GNNs (TGAT/TGN), Graph Transformers (GraphGPS/Graphormer), NeRF/instant-NGP, 3D detection, depth completion, Bayesian NNs, conformal prediction, INLA, Neural Audio (Conv-TasNet, vocoder).

4. I/O & Interop
   Delta Lake, TileDB, Arrow Flight, Kafka I/O + OxiCode serialization + Python bindings (PyO3, NumPy interop) + WASM (WebGPU backend, 76 tests, Web Worker parallelism).

Key Rust advantages:

• 100% Pure Rust (zero external system libs required)
• SIMD + GPU acceleration everywhere
• Full no_std + alloc support
• Deterministic builds and reproducible WASM
• Zero-warning policy (Clippy0 + fail0 enforced)

What’s inside 0.4.0 (released March 26)

• Major Pure Rust migration completed (OxiBLAS / OxiFFT / OxiARC now default)
• New AI modules: Flash Attention 2, LoRA family, ONNX export, Temporal GNNs, NeRF/instant-NGP
• GPU acceleration: PDE solvers, FFT pipeline, SpMV
• Expanded I/O: Delta Lake, TileDB, Arrow Flight, Kafka
• WASM: Full WebGPU backend with 76 tests
• Docs: mdBook (26 EN + 12 JP pages)
• Production readiness confirmed through 25,800+ passing tests (36,475 #[test] annotations) across 7,640 files
• 2.91 million SLoC (2,908,818 lines) of pure Rust — zero warnings

This is the foundation

SciRS2 is now the official scientific computing and AI backend for the entire COOLJAPAN stack (total ecosystem: 21M+ SLoC Rust, 597 crates, 40+ production-grade libraries):

• OxiBLAS / OxiFFT — linear algebra and signal core
• OptiRS — all training optimizers
• OxiMedia — vision and media pipelines
• Spintronics / OxiHuman — physics and biomechanical simulations
• OxiGDAL — geospatial analysis
• ToRSh / OxiRAG — high-throughput tensor and RAG pipelines
• Future integration with OxiLean for formally verified numerics

Repository: https://github.com/cool-japan/scirs

Star the repo if you want high-performance scientific computing and AI without the Python toolchain headaches.

The era of “just pip install numpy scipy” with all its native dependencies is coming to an end.

Pure Rust scientific computing and AI is here — fast, safe, and sovereign.

— KitaSan at COOLJAPAN OÜ March 26, 2026