q27-bonsai-packs

manthedan/q27-bonsai-packs

q27-bonsai-packs is a machine learning model from manthedan. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for q27. The card lists the license as apache-2.0.

Prepacked weights for the q27 inference engine (Apple Silicon Metal). q27 pull downloads and SHA-256-verifies these files; you normally do not fetch them by hand.

Downloads · 30 days

Access

Public

Updated Oct 5, 2026

Repo size

18.1 GB

Likes

Public

Hugging Face

Repo makeup

Click a slice to open those files.

.q2718.1 GB · 100%

At a glance

Library: q27
License: apache-2.0
Access: Public
Created: Jul 20, 2026
Updated: Oct 5, 2026
SHA: 6d59ea13

Base models

prism-ml/Ternary-Bonsai-2-27B-gguf

Model card

From the Hugging Face model README

q27 Bonsai packs

Prepacked weights for the q27 inference engine (Apple Silicon Metal). q27 pull downloads and SHA-256-verifies these files; you normally do not fetch them by hand.

Created using Bonsai by Prism ML.

Bonsai 2 27B, T2 slim (`q27 pull b2`)

file	bytes	SHA-256
`bonsai2-27b-t2-slim.q27`	7,195,341,824	`b85094a0f53c68c25fc73c28e70e05462620417698c7b08a8977d475391f0aeb`
`qwen38-27b-mtp.tok`	7,212,628	`f8adac2001668f3ba68e040e5c13f0982d6a194e2b4d9047cc050fdc6d7c93f2`

A lossless repack of Prism ML's Ternary-Bonsai-2-27B-PQ2_0.gguf into q27's slim T2 container (all 402 ternary tensors keep every code and scale; BF16 GDN gates stored as F16). It is reproducible byte for byte:

python tools/repack.py --bonsai2-container t2 --slim \
  Ternary-Bonsai-2-27B-PQ2_0.gguf bonsai2-27b-t2-slim.q27

The tokenizer is identical to qwen38-27b-mtp.tok in signalnine/Bonsai-2-27B-q27. The smaller T3 pack (q27 pull b2-t3) is served from that upstream repo.

Older packs

bonsai-27b-b1.q27, ternary-bonsai-27b-t2.q27 and their .tok files are first-generation Bonsai packs used by q27 metal-v0.6.x.

License

Library: q27
License: apache-2.0
Created: Jul 20, 2026
Updated: Oct 5, 2026