We have released BitX, a high-performance, lossless compressor as a for safetensors model files (paper,
Rust crate). BitX achieves 10+ GB/s compression throughput, making it practical for large-scale model storage and transfer. We’ve also open-sourced ZipLLM, our model-aware data reduction pipeline. Access the project website to learn more about the project.

Check them out and share your feedback!