MiroPsota/torch_packages_builder

Builder and index for PyTorch packages

What it solves

It simplifies the process of building and distributing PyTorch-based packages that contain custom operators. These packages often require specific hardware architectures and PyTorch versions to work, making them difficult to compile and host for general users.

How it works

The project provides two automated GitHub workflows: one that compiles PyTorch packages with custom ops across various architectures and publishes them as wheels to GitHub releases; another that creates a PEP 503 compliant package index hosted on GitHub Pages, allowing users to install these pre-built binaries via pip install --extra-index-url.

Who it’s for

Developers and researchers who need to install PyTorch extensions with custom operators without having to compile them from source locally.

Highlights

  • Automated building of PyTorch packages with custom ops.
  • PEP 503 compliant package index for easy pip integration.
  • Support for multiple compute platforms including CPU, CUDA, and ROCm.
  • Use of build attestation to establish provenance for wheels.

Related

  • Project
    wildminder/AI-windows-whl

    A curated collection of pre-compiled Python wheels for AI/ML libraries that are difficult to install on Windows, eliminating the need for source compilation.

  • Project
    microsoft/antares

    A compiler solution for PyTorch 2.0 that enables developers to benchmark and optimize custom operators for various hardware accelerators, including DirectX 12 and CUDA.

  • Project
    pytorch/benchmark

    PyTorch Benchmarks is a collection of standardized, lightweight versions of popular deep‑learning models (BERT, ResNet, Stable Diffusion, etc.) that can be run to measure the performance of different PyTorch builds, CUDA versions, and back‑ends. It offers a uniform API, install scripts, and multiple runners (simple test, pytest‑benchmark, custom userbenchmark) plus utilities for low‑noise tuning on AWS g4dn.metal instances.

  • Project
    mlverse/torch

    An R package that brings PyTorch functionality to the R ecosystem, enabling tensor operations and automatic differentiation for machine learning.

  • Dispatch
    Hugging Face kernel-builder: A Guide to Building and Scaling Production-Ready CUDA Kernels

    Hugging Face introduces the kernel-builder library to simplify the development, multi-architecture compilation, and distribution of production-ready CUDA kernels via the Hugging Face Hub.