sarashina2.2 vision 3b needs about 3.9 GB of memory at Q4_K_M (a 2.1 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 7 quants, from 2.0 GB (Q4_0) to 6.7 GB (F16).
Sarashina2.2 Vision 3B – 3.36B parameter vision‑enabled LLaMA model
Sarashina2.2 Vision 3B is a 3.36B parameter LLaMA model with vision capabilities. It is available in GGUF format for local inference via llama.cpp. The model was converted from the original safetensors weights without architectural changes.
Good at
- Vision‑text multimodal tasks
- Local inference with llama.cpp
- Compact quantized variants
Limits
- AGPL‑3.0 license restricts commercial use
- No vision support in quantized variants without mmproj
- Requires 8 k context for full functionality
Best for
- Image captioning on local hardware
- Multimodal chatbots
- Rapid prototyping of vision models
Who it’s for. Developers and researchers needing a lightweight vision‑enabled LLaMA model for local deployment. The model is suitable for projects that require multimodal inference without cloud dependencies.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q4_0 | 2.0 GB | 3.8 GB | Fits | Get file → | |
| Q4_K_M | 2.1 GB | 3.9 GB | Fits | Get file → | |
| Q4_K_M | 2.1 GB | 3.9 GB | Fits | Get file → | |
| Q8_0 | 3.6 GB | 5.4 GB | Fits | Get file → | |
| — | 6.7 GB | 8.6 GB | Too big | Get file → | |
| BF16 | 6.7 GB | 8.6 GB | Too big | Get file → | |
| F16 | 6.7 GB | 8.6 GB | Too big | Get file → |
For 8 GB VRAM, use the Q4_0 or Q4_K_M quantized files (~1.96 GB or 2.07 GB). For 12 GB VRAM, the Q8_0 quant (~3.57 GB) offers higher precision. If memory is limited, the Q4_0 variant provides the best balance of size and performance.
What GPU do you need for sarashina2.2 vision 3b?
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 Q4_K_M (2.1 GB file, about 3.9 GB in use).
- 6 GB VRAM: the best quant that fits is Q8_0 (3.6 GB file, about 5.4 GB in use).
- 8 GB VRAM: the best quant that fits is Q8_0 (3.6 GB file, about 5.4 GB in use).
- 12 GB VRAM: the best quant that fits is the main file (6.7 GB file, about 8.6 GB in use).
Which sarashina2.2 vision 3b quant should you choose?
Q4_K_M is the usual starting point for sarashina2.2 vision 3b: it balances file size and quality well. Here it is a 2.1 GB file.
The Q8_0 file is 3.6 GB, about 1.7 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.
If memory is tight, the smallest file is Q4_0 at 2.0 GB, with the quality trade-offs that come with fewer bits. The usual sweet spot between file size and quality, and the quant most people start with.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about sarashina2.2 vision 3b
How much VRAM does sarashina2.2 vision 3b need?
About 3.9 GB at Q4_K_M with an 8K context window. That is the 2.1 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 sarashina2.2 vision 3b on an 8 GB GPU?
Yes, with Q8_0: the 3.6 GB file needs about 5.4 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of sarashina2.2 vision 3b GGUF?
It depends on the quant: Q4_0 is 2.0 GB, Q4_K_M is 2.1 GB, Q4_K_M is 2.1 GB, Q8_0 is 3.6 GB, file is 6.7 GB, BF16 is 6.7 GB and 1 more.
What context length does sarashina2.2 vision 3b support?
The GGUF header lists a trained context length of 8,192 tokens (8K). The KV cache grows with the context you use: about 1.3 GB at 8K and 1.3 GB at 8K.
Where can I download sarashina2.2 vision 3b GGUF?
The files are on Hugging Face in the samuraieng/sarashina2.2-vision-3b-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 sarashina2.2 vision 3b use?
Hugging Face lists the license as agpl-3.0. 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 Sarashina2.2 Vision 3B on 8 GB VRAM?
Yes, the Q4_0 and Q4_K_M quantized files fit within 8 GB VRAM, with estimated memory usage of 3.8 GB and 3.9 GB at 8 k context.
Does the model support vision tasks out of the box?
Yes, the model has detected vision abilities and can process image inputs when used with compatible mmproj files.
What license applies to this model?
The model is released under the AGPL‑3.0 license, which requires source disclosure for derivative works.