Ternary-Bonsai-2-9B-DFlash2 needs about 3.1 GB of memory at its main file (a 2.4 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 5 quants, from 1.6 GB (Q8_0) to 2.7 GB (Q8_0). Longer context needs more: at 32K tokens the KV cache alone adds about 0.8 GB.
Ternary-Bonsai-2-9B-DFlash2: 2.5B vision‑language model with ternary weights
Ternary‑Bonsai‑2‑9B‑DFlash2 is a 2.5B parameter vision‑language model derived from Qwen3.5‑9B. It uses a dflash architecture and supports 262,144‑token context. The model is distributed in GGUF format with several ternary quantizations.
Good at
- Vision‑language inference
- Large context handling
- Low‑memory deployment
Limits
- No text‑only quantization available
- Requires custom llama.cpp build for ternary kernels
- Vision projector must be loaded separately
Best for
- Multimodal chatbots
- Image captioning
- Vision‑guided code generation
Who it’s for. Developers needing a compact vision‑language model that fits on modest GPUs. The model is suitable for edge deployments where memory is limited but a 262K context is required.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q8_0 | 1.6 GB | 2.3 GB | Fits | Get file → | |
| — | 2.0 GB | 2.7 GB | Fits | Get file → | |
| — | 2.4 GB | 3.1 GB | Fits | Get file → | |
| Q2_0 | 2.6 GB | 3.3 GB | Fits | Get file → | |
| Q8_0 | 2.7 GB | 3.4 GB | Fits | Get file → |
For 8 GB VRAM, use the Q8_0 quantization (1.57 GB file, ~2.3 GB memory). If 4 GB VRAM is available, the Q2_0 quantization (2.56 GB file, ~3.3 GB memory) is the smallest option that still runs. The PTQ1_0 pack (2.00 GB) is not listed in the file set, so it cannot be used here.
What GPU do you need for Ternary-Bonsai-2-9B-DFlash2?
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 Q8_0 (2.7 GB file, about 3.4 GB in use).
Which Ternary-Bonsai-2-9B-DFlash2 quant should you choose?
If memory is tight, the smallest file is Q8_0 at 1.6 GB, with the quality trade-offs that come with fewer bits. Close to lossless. It costs roughly twice the memory of a 4-bit file, so it suits cases where you have plenty of VRAM.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Ternary-Bonsai-2-9B-DFlash2
How much VRAM does Ternary-Bonsai-2-9B-DFlash2 need?
About 3.1 GB at its main file with an 8K context window. That is the 2.4 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-2-9B-DFlash2 on an 8 GB GPU?
Yes, with Q8_0: the 2.7 GB file needs about 3.4 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Ternary-Bonsai-2-9B-DFlash2 GGUF?
It depends on the quant: Q8_0 is 1.6 GB, file is 2.0 GB, file is 2.4 GB, Q2_0 is 2.6 GB, Q8_0 is 2.7 GB.
What context length does Ternary-Bonsai-2-9B-DFlash2 support?
The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 0.2 GB at 8K and 0.8 GB at 32K.
Where can I download Ternary-Bonsai-2-9B-DFlash2 GGUF?
The files are on Hugging Face in the prism-ml/Ternary-Bonsai-2-9B-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-2-9B-DFlash2 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 Q8_0 quantization uses 1.57 GB file size and requires about 2.3 GB memory at 8K context, which fits within 8 GB VRAM.
Does the model support vision input?
Yes, vision input is handled by a separate mmproj pack (Q8_0). It must be loaded only when an image is provided.
What license applies to this model?
The model is released under the Apache‑2.0 license.