Driving the Future of Open Source AI: An Update from PyTorch Foundation Projects
Latest updates from PyTorch Foundation and its contributions to AI innovation.
1PyTorch Foundation's New Structure
In April 2025, PyTorch Foundation transitioned to a multi-project structure to scale innovation throughout the AI lifecycle and support cross-domain collaboration. This new structure encompasses six projects.
2Latest Updates from PyTorch
PyTorch 2.13 release brought Apple Silicon optimization and performance enhancements. New features include FlexAttention, CuTeDSL Inductor backend, and Python 3.15 support.
3Achievements of the vLLM Project
The vLLM project stands out with the redesign of Model Runner V2 and significant performance improvements on GPTQ. The project also published its Q3 2026 roadmap.
4DeepSpeed's Innovations
DeepSpeed continues rapid software innovations with major architectural updates like Torch.xpu integration and the Ulysses parallelism algorithm. The project is also recognized in the academic community.
5Ray's Large-Scale Workloads
Ray focuses on production hardening for large-scale workloads with GB200/GB300 hardware support and scalable model optimizations. Ray Data 2.57 version is also set to release soon.
6Helion's First Six Months
Helion continues its efforts to deliver state-of-the-art performance across heterogeneous hardware. The project focuses on LLM-guided autotuning and cross-hardware kernels.
7Importance for the Future of AI
PyTorch Foundation and its projects are shaping the future of the open-source ecosystem by supporting AI innovation. These updates are making AI more accessible to a broader audience.
Original article: PyTorch Blog — PyTorch Foundation Projeleri ile Açık Kaynak AI'nın Geleceği