LLVM Foundation
foundation · United StatesScores
1 product on the map — 1 open.
Openness
5 high confidence- license
- Apache-2.0-WITH-LLVM-exception (OSI, offered alongside a BSD-style PyTorch license at the user's choice)
- source
- public
- core features withheld
- no
Torch-MLIR is licensed under Apache-2.0 with LLVM exceptions, with a BSD-style license offered as an alternative, and the whole project builds from its public repository. It is run as an LLVM incubator project, with no company selling an edition beside it, so nothing is held back from the published source.
- https://pypi.org/pypi/torch-mlir/json recorded 2026-11-12
PyPI metadata for torch-mlir: summary "First-class interop between PyTorch and MLIR", with monthly releases 20261001 and 20261101 uploaded on 2026-10-01 and 2026-11-01.
- https://raw.githubusercontent.com/llvm/torch-mlir/main/LICENSE recorded 2026-11-12
LICENSE body headed "Dual License": "The torch-mlir project is dual licensed under both the standard LLVM Project license of Apache License v2.0 with LLVM Exceptions and a BSD-style license from PyTorch. As a user of this code you may choose to use it under either license."
- https://raw.githubusercontent.com/llvm/torch-mlir/main/README.md recorded 2026-11-12
README: "This project is participating in the LLVM Incubator process: as such, it is not part of any official LLVM release"; it documents installing pre-built snapshots and building from source, and names no paid tier, commercial edition or license-gated build.
Adoption
1 high confidencePyPI downloads of the torch-mlir package, which the project publishes from its own release workflow. Nightly wheels installed from the project's GitHub release page, and copies built into downstream compilers such as IREE, do not appear in the count.
- https://pypi.org/project/torch-mlir/ recorded 2026-11-12
The PyPI project page shows the files were uploaded using Trusted Publishing from llvm/torch-mlir's .github/workflows/release.yml, so the package is the project's own.
- https://pypistats.org/api/packages/torch-mlir/recent recorded 2026-11-12
"last_month":1477 downloads for torch-mlir.
Capability
3 medium confidenceTorch-MLIR takes models from PyTorch and ONNX into MLIR and lowers them to three of the ecosystem's standard dialects, which gives other compilers a shared PyTorch front end. It stops there: the final compilation for a device is left to tools such as IREE, so it sits a step below TensorRT, which compiles and runs the model on NVIDIA hardware.
- https://raw.githubusercontent.com/llvm/torch-mlir/main/docs/architecture.md recorded 2026-11-12
Architecture doc: Torch-MLIR "does not attempt to provide a production end-to-end flow for PyTorch programs by itself"; its backend lowers the backend contract to "Linalg-on-Tensors", "TOSA" and "StableHLO".
- https://raw.githubusercontent.com/llvm/torch-mlir/main/README.md recorded 2026-11-12
README: "Torch-MLIR is primarily a project that is integrated into compilers to bridge them to PyTorch and ONNX", naming IREE and Blade as existing downstreams, with ONNX and FX listed as entry points.