---

license: apache-2.0

library_name: llama.cpp

pipeline_tag: text-generation

tags:

- gguf

- text-generation

- chat

- ternary

- quantized

- on-device

- mirror

language:

- en

base_model: prism-ml/Ternary-Bonsai-27B-gguf

---

Model Card for Ternary Bonsai 27B (GGUF, mirror)

An unmodified mirror of one quant of Prism ML's Ternary Bonsai 27B GGUF. This is

a byte-for-byte copy re-hosted under the Apache 2.0 license as a stable pinned

snapshot that survives if the upstream repo is moved or removed. It is NOT the

authoritative source and adds nothing to the model.

This repository holds a quantized weight file and its matching vision projector

only, not training data or original weights, and nothing here was re-quantized,

converted, or otherwise changed. For the full model card, benchmarks, other

quants, and the whitepaper, use the original repository.

Model Details

Model Description

Ternary Bonsai 27B is a Qwen3.5 hybrid model (general.architecture = qwen35):

of its 64 layers, 48 are Gated DeltaNet linear-attention (state-space) blocks and

16 are full softmax-attention blocks, interleaved "three linear, one attention".

The weights are quantized to a true ternary alphabet {-1, 0, +1} at ~1.71 bits

per weight. This mirror carries the Q2_0_g128 build: each 128-weight block is

{ FP16 scale d; 2-bit codes qs[32] } and dequantizes as w = (code - 1) * d.

• Developed by: Prism ML (model + ternary quantization), built from

Qwen3.6-27B by Alibaba Cloud; this repository is an unmodified mirror by leok7v

• Model type: Hybrid Gated DeltaNet + attention causal language model,

ternary-quantized GGUF (llama.cpp)

• Language(s): English and the languages of the base model
• License: Apache 2.0 (inherited unchanged from the upstream model)
• Mirrored from model:

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

Model Sources

• Repository (this mirror): https://huggingface.co/leok7v/Ternary-Bonsai-27B-gguf
• Original (authoritative) repository:

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

(Apache 2.0)

• Base model:

Qwen/Qwen3.6-27B (Apache 2.0)

Attribution

Per the upstream NOTICE:

> Created using Bonsai by Prism ML.

Copyright 2026-present Prism ML, Inc.; built from Qwen3.6-27B, Copyright 2026

Alibaba Cloud. See LICENSE.txt (Apache 2.0) and NOTICE.txt.

Contents

| File | Notes |

| :--- | :--- |

| Ternary-Bonsai-27B-Q2_0.gguf (~7.2 GB) | Byte-for-byte copy of the upstream Q2_0_g128 ternary build. |

| Ternary-Bonsai-27B-mmproj-Q8_0.gguf (~0.63 GB) | Byte-for-byte copy of the upstream Q8_0 vision projector (mmproj) for the multimodal path. |

Why a mirror and not a fork

Hugging Face does not support forking a model repository, so mirroring the weight

file is the only way to keep a pinned snapshot; the storage and bandwidth are the

mirror's cost, not upstream's. If you just want the model, prefer the original.