uber-research/sbnet
A TensorFlow library of custom CUDA operations that implements Sparse Blocks Networks for efficient neural network inference.
Not currently ranked — collecting fresh signals.
star history
This repository contains a TensorFlow custom operations library implementing SBNet (Sparse Blocks Network) for faster inference, along with a Python implementation of sparse ResNet blocks. It provides custom CUDA kernels for ops like reduce_mask, sparse_gather, and sparse_scatter that work on active block regions rather than dense tensors. The library also benchmarks against Submanifold Sparse Convolutional Networks.
Frequently asked
- What is uber-research/sbnet?
- A TensorFlow library of custom CUDA operations that implements Sparse Blocks Networks for efficient neural network inference.
- Is sbnet open source?
- Yes — uber-research/sbnet is an open-source project tracked on heatdrop.
- What language is sbnet written in?
- uber-research/sbnet is primarily written in Python.
- How popular is sbnet?
- uber-research/sbnet has 433 stars on GitHub.
- Where can I find sbnet?
- uber-research/sbnet is on GitHub at https://github.com/uber-research/sbnet.