OmniVLM 968M needs about 1.1 GB of memory at Q8_0 (a 0.5 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 4 quants, from 0.5 GB (Q8_0) to 1.0 GB (F16). Longer context needs more: at 32K tokens the KV cache alone adds about 0.4 GB.
OmniVLM 968M – a compact vision‑language model for edge devices
OmniVLM 968M is a sub‑billion parameter multimodal model based on qwen2 architecture. It supports vision and tool use with a 131072 token context window. The model is released under Apache‑2.0 license.
Good at
- Vision‑language tasks
- Tool integration
- Edge inference
Limits
- No vision support on CPU only setups
- Requires 8k context for memory estimates
- No large‑model MOE
Best for
- Mobile image captioning
- On‑device visual question answering
- Lightweight multimodal chat
Who it’s for. Developers building multimodal applications on devices with limited memory. The model is suitable for projects needing both image and text understanding without heavy GPU resources.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q8_0 | 0.5 GB | 1.1 GB | Fits | Get file → | |
| Q8_0 | 0.5 GB | 1.1 GB | Fits | Get file → | |
| F16 | 1.0 GB | 1.6 GB | Fits | Get file → | |
| F16 | 1.0 GB | 1.6 GB | Fits | Get file → |
For 8GB VRAM, use Q8_0 (0.53 GB file, 1.1 GB memory at 8k context). For 12GB VRAM, F16 (0.99 GB file, 1.6 GB memory) offers higher precision. Memory estimates are approximate.
What GPU do you need for OmniVLM 968M?
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 F16 (1.0 GB file, about 1.6 GB in use).
Which OmniVLM 968M quant should you choose?
Q8_0 (0.5 GB) is the middle option in this repository. Close to lossless. It costs roughly twice the memory of a 4-bit file, so it suits cases where you have plenty of VRAM.
If memory is tight, the smallest file is Q8_0 at 0.5 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 OmniVLM 968M
How much VRAM does OmniVLM 968M need?
About 1.1 GB at Q8_0 with an 8K context window. That is the 0.5 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 OmniVLM 968M on an 8 GB GPU?
Yes, with F16: the 1.0 GB file needs about 1.6 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of OmniVLM 968M GGUF?
It depends on the quant: Q8_0 is 0.5 GB, Q8_0 is 0.5 GB, F16 is 1.0 GB, F16 is 1.0 GB.
What context length does OmniVLM 968M support?
The GGUF header lists a trained context length of 131,072 tokens (128K). The KV cache grows with the context you use: about 0.1 GB at 8K and 0.4 GB at 32K.
Where can I download OmniVLM 968M GGUF?
The files are on Hugging Face in the NexaAI/OmniVLM-968M repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does OmniVLM 968M 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 OmniVLM 968M on 8 GB VRAM?
Yes, the Q8_0 quant uses about 1.1 GB memory at 8k context, fitting within 8 GB VRAM.
Does OmniVLM support vision tasks?
Yes, OmniVLM is designed for vision‑language tasks and includes vision capabilities.
What license does OmniVLM use?
OmniVLM is released under the Apache‑2.0 license.