Roadmap

v1.0-draft spec implemented. Static shapes & dtypes.

Core v1 conformance is defined and tested; see Conformance and Stability.

The foundational language specification with type system and core syntax.

Reverse-mode AD for all Core v1 ops: Add, Sub, Mul, Div, MatMul, Dot, Transpose, Sum, Mean, Reshape, Relu, Conv2d, and indexing/slicing operations.

Full gradient support for Core v1 operations including Conv2d.

Comprehensive benchmarks verified: ~38 µs compilation (53-247× faster than PyTorch 2.0), 100% bit-level determinism, 1,345-11,284× more efficient autodiff.

See Performance and Running Benchmarks for full details.

Python bindings, determinism verification, PyTorch comparison, and documentation complete.

Deterministic CPU backend via the MIND runtime interface (Phase-1 ops).

Production-ready CPU execution with optimized compilation.

All 4 GPU backends production-ready: CUDA (NVIDIA), Metal (Apple), ROCm/HIP (AMD), and WebGPU (cross-platform). Full trait implementations with poison-safe locking, defragmentation, and context recovery.

Enterprise runtime achieves 35% faster matmul (TF32), 40% faster FP16, 98% memory bandwidth. Available via Enterprise license. All backends implement StreamSync, ContextRecovery, and Defragmenter traits.

Production GPU backends for NVIDIA, AMD, Apple Silicon, and browsers/native via WebGPU.

Early groundwork for module and dependency resolution with ecosystem foundations underway.

Comprehensive package management and dependency resolution.

Ultra-low latency runtime paths, streaming tensor support, and signal processing primitives for brain-computer interface and neuroscience applications.

Targeting <1ms inference latency, pre-allocated memory pools, and deterministic timing guarantees for medical device certification.

Initial research and design phase. Runtime optimizations, signal processing stdlib, and language extensions planned for late 2026.

Multi-node training and inference with automatic model sharding, data parallelism, and collective communication (NCCL, Gloo).

See Distributed Execution Guide for early access documentation.

Full-stack distributed training with pipeline parallelism and elastic scaling.

Production-ready model serving with containerized deployment, auto-scaling, A/B testing, and model versioning.

See Deployment Guide for deployment options and best practices.

One-command deployment to cloud, edge, and on-premise environments.

Spec aligned with implementations. 20 IR instructions, 47 E-codes, 1e-5 numerical tolerance. 154 compiler tests, 37 runtime tests, all passing.

Compiler and runtime implementations fully match spec. Conformance test suite complete.

All spec files aligned: ir.md, autodiff.md, ffi.md, runtime.md, types.md, errors.md. CHANGELOG.md present in all repos.

Specification-grade documentation across compiler, runtime, and spec repos.

5 MLIR dialects (arith, tensor, linalg, func, scf). Tested with LLVM 18. 7 FFI functions, 3 enums, 2 opaque types.

Full compiler workflow. GPU/accelerator extensions in progress.