A transposed coordinate list could move values to the wrong row in CSR and BSR conversion, and the CPU path could write past its output.
By [RuntimeWire Staff](https://runtimewire.com/author/runtimewire-staff)
· Published
PyTorch has merged a fix for a sparse-tensor bug that returned wrong results without raising an error and, on CPU, could write outside its output buffer and corrupt the heap. The merged commit landed on the main branch on November 15th, 2026, and changes the CPU and CUDA paths that convert COO indices to CSR indices.…