ZebraLSTM is an advanced Long Short-Term Memory (LSTM) architecture designed to optimize the performance of sequential data processing tasks. By dividing the hidden state into multiple parallel stripes, ZebraLSTM enhances computational efficiency and scalability, particularly when leveraging modern GPU architectures.
- Striped hidden state processing for parallelism and efficiency
- Configurable architecture for various applications
- Suitable for NLP, time-series analysis, and drug design
src/: Source code for ZebraLSTMexamples/: Example scripts and notebookstests/: Unit tests and integration testsdocs/: Documentation and referencescuda/: CUDA implementations and related scriptsscripts/: Utility scripts for setup and installation
To install dependencies, run:
pip install -r requirements.txt
Copyright 2024 R Gagnon