Get the latest tech news
Compress any LLM 3x in one click,near-lossless,in under 50 MB RAM
The peer-reviewed publication of SPECTRA (Zenodo DOI: 10.5281/zenodo.24058202). 3.31x model compression, calibration-free Group-32 microscaling, and 82.53 tok/s decoding on 4GB consumer GPUs.
None
Or read this on Hacker News