为PyTorch补齐了可微稀疏线性代数能力，支持多设备、多后端高效求解。
torch-sla: Differentiable Sparse Linear Algebra with Adjoint Solvers and Sparse Tensor Parallelism for PyTorch
- 统一自动微分接口，自动按设备和问题规模切换求解器后端。
- 支持批量求解与共享/独立稀疏模式，可在多GPU间分布式执行。
- 采用O(1)图反向传播框架，实现稀疏矩阵求解的高效可微计算。
可微稀疏线性代数是科学机器学习的基础，但PyTorch缺乏统一库：torch.sparse仅提供底层内核，且spsolve非可微、仅限CPU；torch.linalg则仅支持稠密运算。本文提出torch-sla，一个开源库，填补此空白。它通过单一自动微分感知接口，支持直接、迭代、非线性及特征值求解器，兼容五种后端——CPU上的SciPy与Eigen，GPU上的cuDSS、CuPy及原生PyTorch迭代求解器——并根据设备与问题规模自动调度。支持基于共享或独立稀疏模式的批量求解，以及通过域分解与环带交换实现的多GPU分布式执行。其可扩展性得益于O(1)图反向传播框架与自动微分兼容的分布式环带交换层。项目地址：https://www.torchsla.com/。
原文摘要 · Abstract (English)
Differentiable sparse linear algebra is foundational for scientific machine learning, yet PyTorch lacks a unified library for it: torch.sparse provides only low-level kernels and a non-differentiable, CPU-only spsolve, and torch.linalg is dense-only. We present torch-sla, an open-source library that fills this gap. It exposes a single autograd-aware API for direct, iterative, nonlinear, and eigenvalue solvers across five interchangeable backends -- SciPy and Eigen on CPU, cuDSS, CuPy, and a PyTorch-native iterative solver on GPU -- with automatic dispatch by device and problem size. The library further supports batched solves over shared or distinct sparsity patterns and distributed multi-GPU execution via domain decomposition with halo exchange. These capabilities are made scalable by an O(1)-graph adjoint differentiation framework and an autograd-compatible distributed halo-exchange layer. The library is available at https://www.torchsla.com/.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。