torch-sla: Differentiable Sparse Linear Algebra with Adjoint Solvers and Sparse Tensor Parallelism for PyTorch
📰 ArXiv cs.AI
Learn to use torch-sla for differentiable sparse linear algebra in PyTorch, enabling scientific machine learning applications
Action Steps
- Install torch-sla using pip to enable differentiable sparse linear algebra in PyTorch
- Use torch-sla's autograd-aware API to solve sparse linear systems with direct, iterative, nonlinear, and eigenvalue solvers
- Apply torch-sla to scientific machine learning applications, such as physics-informed neural networks and sparse regression
- Configure torch-sla to utilize sparse tensor parallelism for improved performance
- Test and validate torch-sla's performance on various sparse linear algebra problems
Who Needs to Know This
Data scientists and machine learning engineers working with PyTorch can leverage torch-sla to improve performance and accuracy in scientific machine learning tasks, such as solving sparse linear systems and eigenvalue problems
Key Insight
💡 torch-sla fills the gap in PyTorch's support for differentiable sparse linear algebra, enabling more accurate and efficient scientific machine learning applications
Share This
✅ Introducing torch-sla: a unified library for differentiable sparse linear algebra in PyTorch! ✅
Key Takeaways
Learn to use torch-sla for differentiable sparse linear algebra in PyTorch, enabling scientific machine learning applications
Full Article
Title: torch-sla: Differentiable Sparse Linear Algebra with Adjoint Solvers and Sparse Tensor Parallelism for PyTorch
Abstract:
arXiv:2601.13994v2 Announce Type: replace-cross Abstract: Differentiable sparse linear algebra is foundational for scientific machine learning, yet PyTorch lacks a unified library for it: \texttt{torch.sparse} provides only low-level kernels and a non-differentiable, CPU-only \texttt{spsolve}, and \texttt{torch.linalg} is dense-only. We present \torchsla{}, an open-source library that fills this gap. It exposes a single autograd-aware API for direct, iterative, nonlinear, and eigenvalue solvers
Abstract:
arXiv:2601.13994v2 Announce Type: replace-cross Abstract: Differentiable sparse linear algebra is foundational for scientific machine learning, yet PyTorch lacks a unified library for it: \texttt{torch.sparse} provides only low-level kernels and a non-differentiable, CPU-only \texttt{spsolve}, and \texttt{torch.linalg} is dense-only. We present \torchsla{}, an open-source library that fills this gap. It exposes a single autograd-aware API for direct, iterative, nonlinear, and eigenvalue solvers
DeepCamp AI