TensorIR

license

Apache-2.0(OSI)

source

public

core features withheld

no

Apache-2.0 license body confirmed. The repository is public and unarchived and builds the whole product, and the README describes no paid tier, enterprise edition or license-gated build beside it, so source is public and the core ungated.

• https://api.github.com/repos/NVIDIA/tensor-ir recorded 2026-09-02

  Repo metadata for nvidia-tensor-ir's canonical repository - private false, archived false, fork false.

• https://github.com/NVIDIA/tensor-ir/blob/main/LICENSE recorded 2026-10-07

  LICENSE body carries the verbatim Apache License 2.0 text under an NVIDIA copyright header.

• https://github.com/NVIDIA/tensor-ir/blob/main/README.md recorded 2026-10-07

  README says "TensorIR is being released early so the community can evaluate the project" and that "TensorIR Compiler depends on" CUDA Tile IR "for GPU code generation"; it names no paid tier, enterprise edition or license-gated build beside the published source.

No package registry publishes TensorIR, and the repository is built from source, so GitHub stars are the only adoption signal available. A star is not a use.

• https://api.github.com/repos/NVIDIA/tensor-ir recorded 2026-09-02

  Repo metadata - stargazers_count = 111 - for NVIDIA/tensor-ir.

TensorIR is a compiler frontend that lowers tensor computations through CUDA Tile IR to NVIDIA GPU code. That toolchain is narrowed to one vendor's targets, one step below Apache TVM's multi-vendor reach.

• https://github.com/NVIDIA/tensor-ir/blob/main/README.md recorded 2026-10-07

  README says "TensorIR is being released early so the community can evaluate the project" and that "TensorIR Compiler depends on" CUDA Tile IR "for GPU code generation"; it names no paid tier, enterprise edition or license-gated build beside the published source.