torch-sla: Differentiable Sparse Linear Algebra with Adjoint Solvers and Sparse Tensor Parallelism for PyTorch
Researchers have developed torch-sla, an open-source Python library designed to provide differentiable sparse linear algebra capabilities within PyTorch. This library addresses a gap in PyTorch's existing functionalities, which currently offer only low-level kernels or CPU-only, non-differentiable solvers. Torch-sla supports a unified API for various solver types across multiple backends, including CPU and GPU options, and enables distributed execution for enhanced scalability. AI
IMPACT Enables more advanced scientific machine learning models by providing essential differentiable sparse linear algebra tools.