NVIDIA Legate Sparse#
Legate Sparse is a Legate library that aims to provide a distributed and accelerated drop-in replacement for the scipy.sparse library on top of the Legate runtime. Legate Sparse interoperates with cuPyNumeric, a distributed and accelerated drop-in replacement for NumPy, to enable writing programs that operate on distributed dense and sparse arrays.
Users can write Legate Sparse programs in Python and run them on a CPU or a GPU and as soon as their problem size increases, they can run them on a cluster of GPUs or CPUs without any changes to their code.
Note
The final release supports a subset of APIs and options from scipy.sparse. CSR and DIA formats are supported for sparse matrices.
Contents: