rusty1s/pytorch_scatter

Health v2 · maintenance only

Adoption not part of the score

1745 stars · 207 forks observed · 2026-08-28

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

A PyTorch extension library providing highly optimized scatter and segment (sparse update) operations with sum, mean, min, and max reductions. It supports CPU and GPU with backward implementations, composite functions like scatter_softmax, and pip wheels for major OS/PyTorch/CUDA combinations.

Use cases

• aggregate tensor values by group index in PyTorch
• perform segment reductions on sorted or CSR-compressed indices
• compute scatter softmax or logsumexp over groups
• speed up graph neural network message aggregation on GPU
• implement sparse update operations missing from core PyTorch

When to choose

• you need fast scatter/segment reductions in PyTorch on CPU or GPU
• you're building GNNs or point cloud models requiring grouped aggregation
• you need differentiable, traceable scatter operations with CUDA support

When to avoid

• you use TensorFlow, JAX, or another framework instead of PyTorch
• your reduction needs are simple enough for built-in torch index ops
• you cannot install CUDA-matched wheels in your environment

Facets

library · maturity stable

machine-learning gpu-computing math machine-learning deep-learning gpu-computing python windows pytorch scatter-operations segment-operations sparse-update cuda tensor-operations linux macos gpu

Member repositories

For agents

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