nf12
nf12 is Lossless compression of BF16 models to 12-bits. Contribute to dphnAI/nf12 development by creating an account on GitHub.. It is ranked #62 on the AI Radar, in AI infra & evals, first seen 12 h ago and shared in 5 posts (23.5k views).
GitHub - dphnAI/nf12: Lossless compression of BF16 models to 12-bits · GitHub Skip to content Navigation Menu Sign in Search / Sign in Sign up Appearance settings You signed in with another tab or window. Reload to refresh your session. You signed out in another tab or window. Reload to refresh your session. You switched accounts on another tab or window. Reload to refresh your session. Dismiss alert {{ message }} Repository navigation nf12 Public Notifications You must be signed in to change notification settings Fork 0 ( 0 ) Star 1 ( 1 ) You must be signed in to star a repository About Lossless compression of BF16 models to 12-bits…
What people said about nf12 on X
Very exciting to finally show this off: lossless compression of BF16 models to 12bits (based on NyoomFloat12 from @togethercompute). A lot of work went into this to make it as efficient to run as possible, since no hardware has native support for this dtype. https://github.com/dphnAI/nf12
— @AlpinDale, 12 h ago · 357 likes · see the post
🤯 炸裂！Qwen 3.0-8B从BF16 → 12位完全无损压缩居然成功了！打脸压缩一定有损的言论！ @AlpinDale 基于 Together AI 的 NyoomFloat12，已中 NeurIPS 2026 实现： 🔹99.7%权重指数高4位全是0111，直接砍掉不存，解码时瞬间还原 🔹仅0.3%异常值用escape buffer精确保存，比特级无损 🔹解码后与原BF16逐比特一致，输出结果完全相同 🔹实测内存：Qwen3-8B省21%（3.2GiB），30B-A3B省24.4%（13.88GiB） 🔹吞吐提升最高+17.8%，还省带宽，速度反而更快 🔹vLLM已有PR支持在线量化，直接可用…
— @NFT_Chen, 8 h ago · 30 likes · see the post
Impression work! This is super hard to get speed improvements with this kind of format. If we can achieve lossless compression and better speed, we will see more and more models using this.
— @bnjmn_marie, 10 h ago · 26 likes · see the post
Alternatives to nf12
- Nous Portal — Great time to try it out
- TensorFold — Setup guide
- Spark-X2.6 — 6 open routing models from 0.3B to 27B, built to make decisions and extract exactly what you need in a single call. 🤖
- AI/ML API — Crawl Reddit with your AI agent >
- OrcaRouter — Claim yours
- together.ai — Together AI
nf12 in numbers
- Rank on the AI Radar: #62 of 10548
- Shared in 5 posts by 4 accounts: @AlpinDale, @NFT_Chen, @bnjmn_marie, @keennay
- 23.5k views on those posts
- First seen 12 h ago, last shared 36 min ago
- Pricing seen by AGK Intelligence: open source
- Market: AI infra & evals
FAQ
What is nf12?
Lossless compression of BF16 models to 12-bits. Contribute to dphnAI/nf12 development by creating an account on GitHub. It was first shared on X 12 h ago and is ranked #62 on the AI Radar.
Is nf12 free?
It is open source.
Who shared nf12?
4 accounts on X, including @AlpinDale, @NFT_Chen, @bnjmn_marie, in 5 posts totalling 23.5k views.
Every post is read and classified by AGK Intelligence: what it is, which market it belongs to, and whether the link is a real tool. 65.2k posts from 5k X accounts over the last 14 days, 10.5k tools, 19 markets. New posts are collected several times an hour and the full ranking is recomputed every hour — last update 2026-11-07 16:40 UTC. Full method.