Tommy He's Tufazip v2 Shrinks 1 GB of Wikipedia to 105.39 MB
Summary
Tommy He and Tufa Labs release tufazip v2, a decoder that expands a 100,958,324-byte archive back into the 1 GB enwik9 Wikipedia text, for a ranked total of 105,388,353 bytes with the 4,430,029-byte decoder. Reconstructing the file takes about 645,748 seconds on exactly two NVIDIA B200 GPUs.
Key Points
- Tommy He's submission data describes the algorithm as an online Transformer initialized from a quantized offline Cycle2 prior.
- Compression takes about 631,548 seconds and excludes offline training time, per the repository's submission table.
- The decoder requires Linux x86-64, GCC, CUDA 13.0, cuDNN 9.20, Python 3.12.3 and uv to run.