PyTorch Ecosystem Landscape Welcomes PhysicsNeMo, Unsloth, ONNX, and KTransformers – PyTorch

The PyTorch Ecosystem Working Group has added four new projects to its official landscape. PhysicsNeMo (NVIDIA) is a PyTorch-based framework for AI physics surrogate models covering CFD, structural mechanics, and climate science. Unsloth is an open source framework for training and fine-tuning 500+ LLMs locally with custom Triton kernels for speed and memory efficiency. ONNX, the vendor-neutral open standard for ML model interoperability, joins as a key bridge between PyTorch training and cross-platform deployment. KTransformers, from Tsinghua University, enables LLM inference in low-VRAM environments via CPU-GPU heterogeneous computing and expert offloading for MoE architectures.