Uni20

Uni20 is a C++23 tensor-network research library combining dense Tensor operations, ordered CPU/BLAS/LAPACK backend dispatch, matrix-free Krylov algorithms, and an epoch-based asynchronous runtime. The API is still in active design, but working Async<Tensor> examples now exercise the complete path from scheduling through backend kernels and structured presentation.

About Uni20 See implemented capabilities, runnable vertical slices, and current boundaries.

Get Started Configure, build, test, and run the current CMake workflow.

Async Runtime Understand epochs, schedulers, tensor aliases, and operation lowering.

Krylov Solvers Review Lanczos, Arnoldi, exponential, and restart behavior.

Core Documentation

• About Uni20
• Contributor guide
• Getting started
• Architecture overview
• Roadmap
• Testing
• Scalar policy
• Build information
• Python bindings

Implemented Vertical Slices

• Tensor operations and Async support
• Kernel dispatch and runtime diagnostics
• Async Tensor kernel authoring
• Generated tensors and reshape
• Presentation and structured reports
• Async task and DAG diagnostics

Runtime And Tensor Work

• Async runtime model
• Async buffers and awaiters
• Async schedulers
• Tensor operations, semantics, and Async support
• Generated tensors and reshape
• Storage kind and location
• Backend dispatch

Linear Algebra And Models

• Krylov algorithms
• Krylov precision validation
• Krylov test matrices
• MPS
• MPO operators
• QNum

Development Notes

• Agent-assisted development
• Reviewing changes
• AI guidance
• Trace macros
• Doxygen policy