Skip to content
Hot eventDeveloping

tufazip Ranks Second on Large Text Compression Benchmark

1 reports1 reporting sourcesUpdated 1 h ago

Start here

AI digest

A compressor called tufazip placed second on Matt Mahoney's Large Text Compression Benchmark, compressing the billion-byte Wikipedia file enwik9 to 105,388,353 bytes (105.39 MB), a figure that includes its offline weights and decoder. Its only prediction model is a 12-layer, 163M-parameter Transformer. It starts from a 34M-parameter offline model quantized to 4.33 MB, and two offline-trained layers are copied into twelve layers that keep learning during compression. The decoder repeats the same learning process, so the full model's learned weights never need to be transferred.

Generated from reports · updated 59 min ago

Report timeline

Follow the reports to see the event from different sides.

November 20
  1. Hacker News · AI
    tufazip Takes #2 on the Large Text Compression Benchmark With a 163M-Parameter Transformer

    A compressor called tufazip placed second on Matt Mahoney's Large Text Compression Benchmark, compressing the billion-byte Wikipedia file enwik9 to 105,388,353 bytes (105.39 MB) including its offline weights and decoder. Its sole prediction model is a 12-layer, 163M-parameter Transformer initialized from a 34M-parameter offline model quantized to 4.33 MB, with two offline-trained layers copied into twelve layers that keep learning during compression. The decoder repeats the same learning process, so the full model's learned weights never need to be transferred.

Heat for this event

Not enough continuous observations yet to draw a trend.