llvm/torch-mlir

Health v2 · maintenance only

Flags: no_license

Adoption not part of the score

1892 stars · 728 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Torch-MLIR is a compiler project providing first-class translation of PyTorch programs into the MLIR compiler ecosystem. It lets hardware vendors and compiler authors lower PyTorch models to MLIR dialects and then to custom backends without writing their own PyTorch frontends.

Use cases

• compile pytorch models to mlir
• lower torch programs to custom hardware backends
• build a compiler backend for pytorch without writing a frontend
• convert pytorch to onnx then to mlir
• run pytorch models on heterogeneous accelerators
• integrate pytorch into an mlir-based toolchain

When to choose

• you are a hardware vendor building an MLIR-based backend for PyTorch models
• you need a standardized, community-maintained PyTorch-to-MLIR lowering instead of a custom one
• you are doing compiler research on MLIR dialects for ML workloads

When to avoid

• you just want to deploy PyTorch models to standard runtimes like ONNX Runtime or TensorRT
• you need a production-endorsed LLVM component, since the project is still in LLVM incubation
• your target stack has no MLIR involvement at all

Facets

library · maturity active

compiler machine-learning llm-training machine-learning compilers developer-tools gpu-computing python cpp windows mlir pytorch compiler-infrastructure hardware-lowering llvm-incubator linux macos

Member repositories

For agents

markdown · JSON · MCP: product_card(name="llvm/torch-mlir")