Ternary Bonsai 2 27B Uncensored Heretic v2 needs about 8.6 GB of memory at its main file (a 5.9 GB file) with an 8K context window, so it fits on a 12 GB GPU. This repository has 2 quants, from 5.9 GB to 7.2 GB. Longer context needs more: at 32K tokens the KV cache alone adds about 8.6 GB.
Ternary Bonsai 2 27B Uncensored Heretic v2 – 27B ternary transformer with 262K‑token context
This model is a 27‑billion‑parameter ternary transformer built on the Qwen3.8‑27B backbone. It supports 262,144‑token context and includes a separate vision tower. The model is available in two GGUF packings: PTQ1_0 (5.95 GB) and PQ2_0 (7.21 GB).
Good at
- Large context reasoning
- Ternary weight efficiency
- Vision support via separate mmproj
Limits
- Requires Prism‑ML llama.cpp fork for ternary kernels
- Uncensored version has reduced safety alignment
- No built‑in vision evaluation for this release
Best for
- Research on low‑bit ternary models
- Testing 262K‑token context workloads
- Exploring vision‑language integration
Who it’s for. Researchers and developers interested in low‑bit ternary transformer models. The model is intended for experimentation and safety research, not for public deployment.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| — | 5.9 GB | 8.6 GB | Too big | Get file → | |
| — | 7.2 GB | 9.9 GB | Too big | Get file → |
For 8 GB VRAM, use the PTQ1_0 pack (5.95 GB) which estimates 8.6 GB memory at 8 k context. If more VRAM is available, the PQ2_0 pack (7.21 GB) uses 9.9 GB memory at 8 k context and provides a slightly larger weight set. Both packs are consumed directly by llama.cpp without expansion.
What GPU do you need for Ternary Bonsai 2 27B Uncensored Heretic v2?
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 (7.2 GB file, about 9.9 GB in use).
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Ternary Bonsai 2 27B Uncensored Heretic v2
How much VRAM does Ternary Bonsai 2 27B Uncensored Heretic v2 need?
About 8.6 GB at its main file with an 8K context window. That is the 5.9 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 Ternary Bonsai 2 27B Uncensored Heretic v2 on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 8.6 GB, so you would need a larger GPU or CPU offloading.
What is the file size of Ternary Bonsai 2 27B Uncensored Heretic v2 GGUF?
It depends on the quant: file is 5.9 GB, file is 7.2 GB.
What context length does Ternary Bonsai 2 27B Uncensored Heretic v2 support?
The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 2.1 GB at 8K and 8.6 GB at 32K.
Where can I download Ternary Bonsai 2 27B Uncensored Heretic v2 GGUF?
The files are on Hugging Face in the OS-Software/Ternary-Bonsai-2-27B-Uncensored-Heretic-v2-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does Ternary Bonsai 2 27B Uncensored Heretic v2 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 this model on 8 GB VRAM?
Yes, the PTQ1_0 pack is 5.95 GB and estimates about 8.6 GB memory usage at 8 k context, which fits within 8 GB VRAM with some headroom.
Does this model support vision?
Vision support is provided by a separate Q8_0 mmproj pack; the model itself does not include vision weights.
Is this model safe for production use?
No. The uncensored version has reduced safety alignment and is intended only for research and experimentation.