Ternary-Bonsai-2-4B needs about 2.8 GB of memory at Q2_0 (a 1.2 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 5 quants, from 0.9 GB to 1.6 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 4.3 GB.
Ternary-Bonsai-2-4B: 4B vision‑language model with ternary weights
Ternary-Bonsai-2-4B is a 4B‑class vision‑language model based on Qwen3.5‑4B. It uses ternary transformer weights and supports 262K‑token context. The model is available in several GGUF packings and can be run with llama.cpp.
Good at
- Reasoning
- Tools usage
- Vision understanding
Limits
- No vision support without mmproj
- Large memory for high‑precision quant
- License requires Apache‑2.0 compliance
Best for
- Low‑memory inference on CPU
- Multimodal chat with images
- Embedded vision tasks
Who it’s for. Developers needing a compact vision‑language model for edge or CPU deployment. The model is suitable for projects that can use llama.cpp and require a small file size while maintaining reasoning and vision capabilities.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| — | 0.9 GB | 2.5 GB | Fits | Get file → | |
| — | 1.1 GB | 2.7 GB | Fits | Get file → | |
| Q2_0 | 1.2 GB | 2.8 GB | Fits | Get file → | |
| Q8_0 | 1.3 GB | 2.8 GB | Fits | Get file → | |
| BF16 | 1.6 GB | 3.1 GB | Fits | Get file → |
For 8 GB VRAM, use the PTQ1_0 pack (0.95 GB) or PQ2_0 (1.15 GB). If more memory is available, Q2_0 (1.21 GB) or Q8_0 (1.27 GB) provide higher precision. BF16 (1.58 GB) gives the best quality but needs the most memory.
What GPU do you need for Ternary-Bonsai-2-4B?
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: the best quant that fits is BF16 (1.6 GB file, about 3.1 GB in use).
Which Ternary-Bonsai-2-4B quant should you choose?
Q2_0 (1.2 GB) is the middle option in this repository. Use it only when memory is very tight. Expect a clear drop in quality, which is usually worse on small models and easier to live with on large ones.
If memory is tight, the smallest file is the first one in the table at 0.9 GB, with the quality trade-offs that come with fewer bits.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Ternary-Bonsai-2-4B
How much VRAM does Ternary-Bonsai-2-4B need?
About 2.8 GB at Q2_0 with an 8K context window. That is the 1.2 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-2-4B on an 8 GB GPU?
Yes, with BF16: the 1.6 GB file needs about 3.1 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Ternary-Bonsai-2-4B GGUF?
It depends on the quant: file is 0.9 GB, file is 1.1 GB, Q2_0 is 1.2 GB, Q8_0 is 1.3 GB, BF16 is 1.6 GB.
What context length does Ternary-Bonsai-2-4B support?
The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 1.1 GB at 8K and 4.3 GB at 32K.
Where can I download Ternary-Bonsai-2-4B GGUF?
The files are on Hugging Face in the suryatmodulus/Ternary-Bonsai-2-4B-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-2-4B 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 Ternary-Bonsai-2-4B on 8 GB VRAM?
Yes, the PTQ1_0 and PQ2_0 packings fit within 8 GB VRAM and run with llama.cpp.
Does the model support image input?
Yes, image input is handled by a separate mmproj pack (Q8_0 or BF16) that is loaded only when an image arrives.
What license applies to this model?
The model is released under the Apache‑2.0 license.