January 2026 (2026-01) focused on expanding PyTorch’s ROCm backend capabilities for sparse linear algebra and hardening cross-API compatibility. Key work delivered includes enabling triangular solve operations in ROCm through hipsparseSpSV and hipsparseSpSM, which re-enabled test_sparse_triangular_solve and broaden ROCm coverage for sparse solvers. The effort also addressed critical API alignment between cuSPARSE and hipSPARSE for SpSM solve, improving stability and test reliability across backends and enabling maintainers to confidently run end-to-end sparse pipelines on ROCm. Overall, this work reduces risk in production workloads that rely on sparse triangular solves and positions PyTorch to exploit ROCm-enabled hardware more effectively.