TE

bonsai 2 27b mtp

TernaryGGUF by decent-jawfishUpdated 2026-09-18
Download on Hugging Face
Parameters27.3B
Context256K
Layers65
Architectureqwen35
Smallest quant7.6 GB
Quick answer

bonsai 2 27b mtp needs about 10.2 GB of memory at its main file (a 7.6 GB file) with an 8K context window, so it fits on a 12 GB GPU. Longer context needs more: at 32K tokens the KV cache alone adds about 8.7 GB.

Bonsai 2 27B MTP – ternary transformer for on‑device reasoning

Bonsai 2 27B MTP is a 27.3B‑parameter model quantized to ternary weights. It offers 262144‑token context and retains 98.2% of FP16 intelligence. The model runs on llama.cpp with CUDA, Metal, or CPU backends.

Good at

  • Full 27B reasoning on a single GPU
  • 262K‑token context on device
  • Multi‑token‑prediction head for draft‑mtp

Limits

  • No vision support – vision tower is separate
  • Requires patched llama.cpp for draft‑mtp
  • Large memory requirement – ~10.2 GB at 8k context

Best for

  • On‑device reasoning tasks
  • Long‑form content generation
  • Speculative decoding with draft‑mtp

Who it’s for. Developers and researchers needing a large‑scale reasoning model that fits on a single GPU or Apple Silicon. The model is suitable for applications that require extended context and high fidelity to FP16 performance.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
—7.6 GB 10.2 GB
Too big Get file →

The only available GGUF pack is the unknown quant with file size 7.56 GB. It requires about 10.2 GB of VRAM at 8k context. For 8 GB VRAM, the model cannot run without further quantization or model splitting.

What GPU do you need for bonsai 2 27b mtp?

These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.

  • 4 GB VRAM: no quant fits at an 8K context.
  • 6 GB VRAM: no quant fits at an 8K context.
  • 8 GB VRAM: no quant fits at an 8K context.
  • 12 GB VRAM: the best quant that fits is the main file (7.6 GB file, about 10.2 GB in use).

Specifications

Architecture from the GGUF header

Architectureqwen35
Parameters27.3B
Layers65
Trained context262,144 tokens
KV cache (fp16)260 KB per token

Source

Quantized bydecent-jawfish
FamilyTernary
Licenseapache-2.0
Downloads (HF)2,295
Last updated2026-09-18

Frequently asked questions about bonsai 2 27b mtp

How much VRAM does bonsai 2 27b mtp need?

About 10.2 GB at its main file with an 8K context window. That is the 7.6 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.

Can I run bonsai 2 27b mtp on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 10.2 GB, so you would need a larger GPU or CPU offloading.

What is the file size of bonsai 2 27b mtp GGUF?

It depends on the quant: file is 7.6 GB.

What context length does bonsai 2 27b mtp support?

The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 2.2 GB at 8K and 8.7 GB at 32K.

Where can I download bonsai 2 27b mtp GGUF?

The files are on Hugging Face in the decent-jawfish/bonsai-2-27b-mtp repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.

What license does bonsai 2 27b mtp use?

Hugging Face lists the license as apache-2.0. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.

Can I run Bonsai 2 27B MTP on 8 GB VRAM?

No. The model requires approximately 10.2 GB of VRAM at 8k context, exceeding an 8 GB limit.

Does the model support vision tasks?

No. The vision tower ships as a separate Q8_0 mmproj pack and is not included in this GGUF.

What license applies to this model?

The model is released under the apache-2.0 license.

Other GGUFs of Ternary-Bonsai-2-27B-gguf

Similar models

Browse all →

More Ternary models

All Ternary →

Other medium models

All medium models →