Phi 3.5 Vision Instruct needs about 7.8 GB of memory at Q8_0 (a 4.1 GB file) with an 8K context window, so it fits on a 8 GB GPU. This repository has 2 quants, from 4.1 GB (Q8_0) to 7.6 GB (F16). Longer context needs more: at 32K tokens the KV cache alone adds about 12.9 GB.
Phi 3.5 Vision Instruct – 3.82B parameter vision model in GGUF format
Phi 3.5 Vision Instruct is a 3.82B parameter model designed for vision tasks. It is available in GGUF format and supports a 131072 token context window. The model is licensed under MIT.
Good at
- Vision instruction following
- Large context handling
- Open source usage
Limits
- No vision support in main branch of llama.cpp
- Requires 8k context for memory estimates
- No other language support listed
Best for
- Image captioning with instructions
- Multimodal Q&A
- Vision-based instruction following
Who it’s for. Developers building multimodal applications who need a lightweight vision model. The model is suitable for projects that can integrate with llama.cpp and require a 131072 token context window.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q8_0 | 4.1 GB | 7.8 GB | Tight | Get file → | |
| F16 | 7.6 GB | 11.4 GB | Too big | Get file → |
For 8 GB VRAM, use the Q8_0 quant with a 4.06 GB file size; it estimates 7.8 GB memory at 8k context. For larger memory budgets, the F16 quant (7.64 GB file size) estimates 11.4 GB memory at 8k context. Estimates are approximate.
What GPU do you need for Phi 3.5 Vision Instruct?
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: the best quant that fits is Q8_0 (4.1 GB file, about 7.8 GB in use).
- 12 GB VRAM: the best quant that fits is F16 (7.6 GB file, about 11.4 GB in use).
Which Phi 3.5 Vision Instruct quant should you choose?
Q8_0 (4.1 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.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Phi 3.5 Vision Instruct
How much VRAM does Phi 3.5 Vision Instruct need?
About 7.8 GB at Q8_0 with an 8K context window. That is the 4.1 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 Phi 3.5 Vision Instruct on an 8 GB GPU?
Yes, with Q8_0: the 4.1 GB file needs about 7.8 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Phi 3.5 Vision Instruct GGUF?
It depends on the quant: Q8_0 is 4.1 GB, F16 is 7.6 GB.
What context length does Phi 3.5 Vision Instruct support?
The GGUF header lists a trained context length of 131,072 tokens (128K). The KV cache grows with the context you use: about 3.2 GB at 8K and 12.9 GB at 32K.
Where can I download Phi 3.5 Vision Instruct GGUF?
The files are on Hugging Face in the abetlen/Phi-3.5-vision-instruct-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 Phi 3.5 Vision Instruct use?
Hugging Face lists the license as mit. 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 Phi 3.5 Vision Instruct on 8 GB VRAM?
Yes, the Q8_0 quant uses about 7.8 GB memory at 8k context, fitting within 8 GB VRAM.
Does this model support text-only tasks?
The model is primarily vision-focused; no explicit text-only capabilities are listed.
Is the model compatible with the latest llama.cpp?
The repo notes it does not currently work with the main branch of llama.cpp.