deep learning framework / tensor computation
PyTorch is a Python deep learning framework offering GPU-accelerated tensor computation and dynamic neural networks via a tape-based autograd system.
Same problem, same approach. Swapping one for another is a config change, not a rewrite.
choose if you need a PyTorch-like framework with different performance/scalability characteristics and recent maintenance.
OneFlow is a deep learning framework designed to be user-friendly, scalable and efficient.
Solves the same problem with a different architecture or at a different layer. Expect to rewrite the integration.
choose if you need a Rust deep learning library with compile-time shape checking, despite its stale maintenance.
Deep learning in Rust, with shape checked tensors and neural networks
choose if you need a lightweight C++ machine learning library with no Python requirement for research or embedded settings.
A C++ standalone library for machine learning
choose if you need a Java/JVM deep learning framework with engine-agnostic backend support.
An Engine-Agnostic Deep Learning Framework in Java
choose if you build on JAX and want a Sonnet-like neural network module library.
JAX-based neural network library
choose if you want a modern Rust deep learning framework with emphasis on flexibility and portability.
Burn is a next generation tensor library and Deep Learning Framework that doesn't compromise on flexibility, efficiency and portability.
choose if you want a Nim tensor library with deep learning support for CPU, GPU, and embedded devices.
A fast, ergonomic and portable tensor library in Nim with a deep learning focus for CPU, GPU and embedded devices via OpenMP, Cuda and OpenCL backends
An earlier or more limited way to do the job. Still the right call when you need something small, proven, or CPU-only.
choose if you want a PyTorch-like framework with Chinese-first docs and are willing to use a codebase that is no longer actively maintained (last push >18 months).
MegEngine 是一个快速、可拓展、易于使用且支持自动求导的深度学习框架
choose for legacy Chainer codebases or if you specifically need the original define-by-run API that predates PyTorch.
A flexible framework of neural networks for deep learning
These projects were analysed and named pytorch among their alternatives. The relationship is not symmetric — how pytorch rates them is a separate judgement, made when pytorch is analysed in its own right.
calls pytorch “choose when you want the largest ecosystem, most active community, and mature tooling in deep learning”
OneFlow is a deep learning framework designed to be user-friendly, scalable and efficient.
calls pytorch “Choose when you need dynamic, imperative programming with strong research community and lower-level control.”
Deep Learning for humans
calls pytorch “when you need a Python-based DL framework with dynamic computation graphs and broader ecosystem than JS.”
A WebGL accelerated JavaScript library for training and deploying ML models.
calls pytorch “Choose when you want the dominant Python framework with a huge ecosystem and dynamic graphs.”
Burn is a next generation tensor library and Deep Learning Framework that doesn't compromise on flexibility, efficiency and portability.