Hacker GitHub

← All repos

z-lab/paroquant

View on GitHub ↗ ★ 344 Python

[ICLR 2026] ParoQuant: Pairwise Rotation Quantization for Efficient Reasoning LLM Inference

4 mentions on Hacker News by 1 person: 0 stories, 4 comments · first 2026-03-23 · latest 2026-03-26

Mentions per year 4 mentions · busiest year 2026 (4)
20082026
AllStoriesComments NewestTopOldest
  1. Newer quantization approaches are even better, 4-bits gets you no meaningful loss relative to FP16: https://github.com/z-lab/paroquant Hopefully Microsoft keeps pushing BitNet too, so only "1.58" bits are needed. I think fractional…
    comment by naasking · 2026-03-26 · in Quantization from the Ground Up
  2. …yes. ParoQuant is hopefully the future here, 4-bit weights with no real degradation from FP16: https://github.com/z-lab/paroquant
    comment by naasking · 2026-03-26 · in Quantization from the Ground Up
  3. …models and long contexts into the range of "possibly runnable" on beefy consumer hardware. [1] https://github.com/z-lab/paroquant
    comment by naasking · 2026-03-25 · in TurboQuant: Redefining AI efficiency with extreme compression
  4. …and eliminate basically all quantization errors relative to fp16/bf16 even going down to 4-bits: https://github.com/z-lab/paroquant
    comment by naasking · 2026-03-23 · in Apply video compression on KV cache to 10,000x less error at Q4 quant