CoreNet: A library for training deep neural networks (github.com)

Article summary

CoreNet is a deep neural network toolkit developed by Apple for training standard and novel models for various tasks, including foundation models, object classification, and semantic segmentation. The library is built on top of PyTorch and allows researchers and engineers to train models efficiently. CoreNet evolved from CVNets, a high-performance library for computer vision, to encompass a broader range of applications. The library provides a range of features, including training recipes, model implementations, and datasets.

Main themes

• Deep Neural Networks
• CoreNet Library
• Apple Development
• PyTorch
• Computer Vision
• Machine Learning

What commenters say

• Homebrew is the de facto standard terminal packaging tool for macOS, and it's not surprising that Apple engineers use it.
• The lack of a native package manager from Apple is a long-standing issue, and some developers feel that the company should release an official binary repository.
• CoreNet's support for training on Apple silicon is unclear, and some commenters question the usefulness of this capability for large-scale training.
• Apple silicon devices can be useful for training certain models, and some users have reported good performance with their MacBooks.
• The usefulness of CoreNet for training models depends on the specific use case and the optimization of CUDA kernels for the workload.
• Some developers prefer MacPorts over Homebrew due to its consistent and managed collection of tools with self-contained dependencies.
• The introduction of a native package manager from Apple could potentially disrupt the existing ecosystem and impact developers who rely on third-party solutions.
• CoreNet's release is seen as a positive development for Apple developers, particularly those working on iPhone or iPad projects.