HaiAI123

Curated Global AI Tools Directory

Industry update
Flat

LookAtMyQuant open-sources compiler that quantizes Qwen3.5-4B to 1.23 GB

11/22/2026 — 11/22, 04:16·1 sources (all)·1 reports (all)

Key takeaways

  • LookAtMyQuant is a quantization compiler that assigns codewords by measuring how compressing each stored parameter affects model outputs, mi…
  • the recipe and code are open-sourced.

Background & analysis

On November 22, 2026, LookAtMyQuant, a quantization compiler, surfaced on the Hacker News front page. The thread, which drew 143 points and 16 comments, points to the project repository at https://github.com/ShinMegamiBoson/lookatmyquant . According to the project description, LookAtMyQuant works by measuring how much each stored parameter of a model degrades the output when it is compressed, then assigning codewords accordingly so the whole model fits within a given size budget with the least possible damage. The material states that on Gemma 4 E2B, its GGUF file achieves a KL score 12x lower than Q4_K_M at the same file size. The method and all of its results are published on the project site, and the code is released under the MIT license.

On Qwen3.5-4B, LookAtMyQuant 2 compresses the model down to 1.23 GB, or 2.3 bits per weight, and the resulting file solves 330 of 540 exact-answer math problems. The comparison figures given in the material are that Prism ML's Bonsai 2 Ternary 4B solves 328 problems at 1.20 GB, and the unquantized original solves 359. The compressed file's KL divergence from the original, measured on held-out text, is reported as 0.45 nats.

At this point the coverage is not fully consistent. The summary says the recipe and code are both open source, while the body text presents the Qwen3.5-4B results under the name LookAtMyQuant 2. The material does not clarify how the open-source scope maps to the different version names, so the open-source status and version alignment remain unclear. The event has not progressed beyond the details disclosed in this single report.

AI-generated from 1 reports · updated 7 hours ago

Related AI tools

updream
24-hour heatHeat index 78 · peak 97 · 7h ago
24 hours agonow

Reports on this story headlines open the original

Today
  1. LookAtMyQuant is a quantization compiler that measures how much each stored parameter hurts output when compressed, then picks codes to fit a size budget with minimal damage. On Qwen3.5-4B, it compresses the model to 1.23 GB at 2.3 bits per weight and solves 330 of 540 exact-answer math problems, with the recipe and code released under MIT.

    Hacker News · 热榜AI score 53

Other stories people are talking about

How is heat calculated?About the method

Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.

This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.

Surge
Discussion rising fast
New
First report within 6 hours
Rising
Still gathering discussion

Back to AI industry updates →