> Canonical page: https://posttrainllm.com/inspiration/bonsai-2

A note of thanks · Packaging

# Bonsai 2

Bonsai sharpened our habit of separating weight-file size from peak runtime memory and capability retained. A fair same-Mac comparison remains a proposed exercise.

01 · The idea

## What stayed with us

Ternary representation, packing overhead, activation transforms, kernel support, artifact size, runtime memory, and capability retention can rank a model differently depending on the deployment constraint.

02 · The local translation

## What we did with it

The retained exercise is a same-Mac comparison sheet covering revisions, task gates, peak RAM, time to first token, prefill, decode throughput, and total task latency.

03 · The boundary

## Where the comparison stops

File size is not peak runtime memory, and retention relative to a source model does not establish frontier parity.

Source trail

## Follow the work

- Original project [Bonsai 2 release ↗](https://prismml.com/news/bonsai-2-27b)
- Further reading [MLX model card ↗](https://huggingface.co/prism-ml/Ternary-Bonsai-2-27B-mlx-2bit)
- Our evidence · study only [Bonsai 2 27B ↗](https://posttrainllm.com/studies/bonsai-2-27b)

Independent appreciation. The named projects have not endorsed or affiliated with PostTrainLLM.