Ternary Bonsai 2 27B

Get the best builds by email. No account needed.

Explore

Apps & Tools

Ternary Bonsai 2 27B

A 1.75 bpw ternary compression of Qwen 3.8 27B for fast consumer GPU inference.

bySudo su

Is this your creation?

PlayableOpen source

Description

A highly optimized quantization of Qwen 3.8 27B designed to fit on 8GB and 12GB GPUs. It features a grafted multi-token-prediction (MTP) head and a custom kernel branch that enables inference speeds of over 50 tokens per second on an RTX 3060.

Descriptions, tags, and model credits may be AI-generated or inferred from public sources and can be incomplete or wrong. Learn more.

Details

Why this is interesting: Unlocks substantial performance gains (up to 145% in some cases) by documenting how to use latent architectural…
Attributed to: Sudo su@sudoingX
Built with: Qwen3.8-27B
Links: huggingface.co
View tweet
View GitHub repo
Tags: #llm #quantization #gguf #ternary

Discovered: X
First posted: X · Sep 13, 2026
Listed: Added 47d ago
Build: Strong
Models: Strong

Get the best builds by email. No account needed.

Explore

Apps & Tools