LaunchAI ModelsSeptember 18, 2026

PrismML releases Ternary Bonsai 2 27B, a 5.9 GB Qwen3.8 27B compression

Read original source →huggingface.co

Ternary Bonsai 2 27B compresses Qwen3.8 27B from 53.80 GB in FP16 to 5.93 GB using ternary {-1, 0, +1} weights at 1.76 effective bits per weight, retaining 98.2% of aggregate benchmark performance across 20 benchmarks. It supports 262K-token context and text-and-image input under Apache 2.0, but the GGUFs require Prism's llama.cpp fork.

People · Babak Hassibi

How this story unfolded

8 days · 6 reports · 9 community posts · 15 of 16 shown

  1. Sep 17
  2. Sep 18
  3. Sep 19
  4. Sep 20
  5. Sep 25

More stories today

Open the live feed