dblalock/bolt
C++ library providing 10-200x speedup for matrix/vector operations via lossy compression, including MADDNESS for approximate matrix multiplication.
Not currently ranked — collecting fresh signals.
star history
Bolt compresses vectors of real-valued data and runs mathematical operations directly on the compressed representations, achieving significant space and compute savings. The repository also includes MADDNESS, an algorithm for fast approximate matrix multiplication with zero multiply-adds under reasonable assumptions. Both are primarily aimed at accelerating neural network inference on CPUs.
Frequently asked
- What is dblalock/bolt?
- C++ library providing 10-200x speedup for matrix/vector operations via lossy compression, including MADDNESS for approximate matrix multiplication.
- Is bolt open source?
- Yes — dblalock/bolt is open source, released under the MPL-2.0 license.
- What language is bolt written in?
- dblalock/bolt is primarily written in C++.
- How popular is bolt?
- dblalock/bolt has 2.5k stars on GitHub.
- Where can I find bolt?
- dblalock/bolt is on GitHub at https://github.com/dblalock/bolt.