rusty1s/pytorch_sparse

Health v2 · maintenance only

Adoption not part of the score

1106 stars · 162 forks observed · 2026-09-27

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

A PyTorch extension library providing optimized sparse matrix operations (coalesce, transpose, sparse-dense and sparse-sparse multiplication) with autograd support. Operations work on CPU and GPU across varying data types and use index/value tensors instead of torch.sparse_coo_tensor.

Use cases

• multiply sparse and dense matrices in pytorch with gradients
• speed up sparse matrix ops on gpu for gnn training
• transpose large sparse tensors efficiently
• coalesce sparse tensor indices and values
• avoid building torch.sparse_coo_tensor manually
• sparse matrix multiplication with autograd support

When to choose

• you need differentiable sparse matrix operations in PyTorch on CPU or GPU
• you are building graph neural networks with PyTorch Geometric
• you want optimized CUDA kernels for sparse-dense and sparse-sparse matmul

When to avoid

• you only need basic sparse tensors already covered by native torch.sparse
• you work outside the PyTorch ecosystem (e.g. TensorFlow, JAX, SciPy)
• you need sparse operations on indices with gradients, since index tensors are not differentiable

Facets

library · maturity active

machine-learning gpu-computing math machine-learning deep-learning python windows pytorch sparse-matrices autograd cuda extension-library graph-neural-networks algorithms linux macos gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="rusty1s/pytorch_sparse")