Introduction

tensor.cx is an independent, Apache-2.0 licensed project for exploring tensor execution. A Python API sits above a backend-neutral C++20 core, with CPU, Apple Metal, and an optional CUDA backend.

The project is pre-alpha. It is useful for runtime experiments, studying backend contracts, and testing small tensor workloads. APIs may change.

Start here

1. Build from source with a CPU-only environment.
2. Run your first tensor operation.
3. Read the backend support matrix before enabling a GPU.
4. Try matmul and a small inference pipeline.

A deliberately compact runtime

• Contiguous row-major tensors, with float32 and a smaller int32 operation set.
• Explicit device transfers and synchronous execution.
• Elementwise add/multiply, 2D matmul, axis reductions, and selected activations and normalization operations.
• CPU references for accelerator correctness checks.
• A separate experimental kernel API.

Autograd, model training, asynchronous streams, distributed execution, and broad dtype coverage are outside the current runtime. There is no PyTorch compatibility promise.

Read the implementation

The source repository contains the runtime, tests, and engineering records. These guides are a curated user entry point. PROJECT.md remains the authoritative phase ledger; architecture and validation records stay with the implementation.