Qwen2.5-VL-3B-Instruct needs about 2.7 GB of memory at Q4_K_M (a 1.9 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 30 quants, from 1.0 GB (IQ2_XXS) to 6.2 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 1.2 GB.
Qwen2.5‑VL‑3B‑Instruct – 3.1B multimodal model in GGUF format
Qwen2.5‑VL‑3B‑Instruct is a 3.1 billion‑parameter vision‑to‑text model. It supports image‑text prompts and can be run with llama.cpp using a fork that enables vision. The model is available in multiple quantizations ranging from IQ2_XXS to Q8_0.
Good at
- Image‑text generation
- Multimodal instruction following
- Low‑bit inference
Limits
- No vision support in main llama.cpp branch
- Requires forked llama.cpp for vision
- License not specified
Best for
- Rapid prototyping of multimodal apps
- Running on GPUs with limited VRAM
- Testing vision‑to‑text pipelines
Who it’s for. Developers and researchers who need a lightweight multimodal model for experimentation. The model is suitable for local inference on consumer hardware with a compatible llama.cpp fork.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| IQ2_XXS | 1.0 GB | 1.8 GB | Fits | Get file → | |
| IQ2_XS | 1.1 GB | 1.9 GB | Fits | Get file → | |
| Q2_K_S | 1.2 GB | 2.0 GB | Fits | Get file → | |
| IQ2_S | 1.2 GB | 2.0 GB | Fits | Get file → | |
| IQ2_M | 1.3 GB | 2.1 GB | Fits | Get file → | |
| IQ3_XXS | 1.3 GB | 2.1 GB | Fits | Get file → | |
| IQ3_XS | 1.4 GB | 2.2 GB | Fits | Get file → | |
| Q3_K_S | 1.4 GB | 2.2 GB | Fits | Get file → | |
| IQ3_S | 1.4 GB | 2.2 GB | Fits | Get file → | |
| IQ3_M | 1.4 GB | 2.2 GB | Fits | Get file → | |
| Q3_K_M | 1.5 GB | 2.4 GB | Fits | Get file → | |
| IQ4_XS | 1.7 GB | 2.5 GB | Fits | Get file → | |
| Q4_0 | 1.7 GB | 2.5 GB | Fits | Get file → | |
| IQ4_NL | 1.8 GB | 2.6 GB | Fits | Get file → | |
| Q4_K_S | 1.8 GB | 2.6 GB | Fits | Get file → | |
| Q4_K_M | 1.9 GB | 2.7 GB | Fits | Get file → | |
| Q4_1 | 1.9 GB | 2.7 GB | Fits | Get file → | |
| Q5_0 | 2.1 GB | 2.9 GB | Fits | Get file → | |
| Q5_K_S | 2.2 GB | 3.0 GB | Fits | Get file → | |
| Q5_K_M | 2.2 GB | 3.0 GB | Fits | Get file → | |
| BF16 | 2.3 GB | 3.1 GB | Fits | Get file → | |
| F16 | 2.3 GB | 3.1 GB | Fits | Get file → | |
| Q5_1 | 2.3 GB | 3.1 GB | Fits | Get file → | |
| Q6_K_M | 2.5 GB | 3.3 GB | Fits | Get file → | |
| BF16 | 2.9 GB | 3.7 GB | Fits | Get file → | |
| F16 | 2.9 GB | 3.7 GB | Fits | Get file → | |
| Q8_0 | 3.3 GB | 4.1 GB | Fits | Get file → | |
| BF16 | 3.6 GB | 4.4 GB | Fits | Get file → | |
| F16 | 3.6 GB | 4.4 GB | Fits | Get file → | |
| BF16 | 6.2 GB | 7.0 GB | Fits | Get file → |
For 8 GB VRAM, use Q4_0 (1.74 GB file, ~2.5 GB memory). For 12 GB VRAM, Q5_0 (2.13 GB file, ~2.9 GB memory) or Q5_K_S (2.17 GB file, ~3 GB memory). Higher precision quants require more VRAM.
What GPU do you need for Qwen2.5-VL-3B-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: the best quant that fits is BF16 (2.9 GB file, about 3.7 GB in use).
- 6 GB VRAM: the best quant that fits is BF16 (3.6 GB file, about 4.4 GB in use).
- 8 GB VRAM: the best quant that fits is BF16 (6.2 GB file, about 7.0 GB in use).
Which Qwen2.5-VL-3B-Instruct quant should you choose?
Q4_K_M is the usual starting point for Qwen2.5-VL-3B-Instruct: it balances file size and quality well. Here it is a 1.9 GB file.
The Q8_0 file is 3.3 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 IQ2_XXS at 1.0 GB, with the quality trade-offs that come with fewer bits. 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.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Qwen2.5-VL-3B-Instruct
How much VRAM does Qwen2.5-VL-3B-Instruct need?
About 2.7 GB at Q4_K_M with an 8K context window. That is the 1.9 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 Qwen2.5-VL-3B-Instruct on an 8 GB GPU?
Yes, with BF16: the 6.2 GB file needs about 7.0 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Qwen2.5-VL-3B-Instruct GGUF?
It depends on the quant: IQ2_XXS is 1.0 GB, IQ2_XS is 1.1 GB, Q2_K_S is 1.2 GB, IQ2_S is 1.2 GB, IQ2_M is 1.3 GB, IQ3_XXS is 1.3 GB and 24 more.
What context length does Qwen2.5-VL-3B-Instruct support?
The GGUF header lists a trained context length of 128,000 tokens (125K). The KV cache grows with the context you use: about 0.3 GB at 8K and 1.2 GB at 32K.
Where can I download Qwen2.5-VL-3B-Instruct GGUF?
The files are on Hugging Face in the Mungert/Qwen2.5-VL-3B-Instruct-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
Can I run Qwen2.5‑VL‑3B‑Instruct on 8 GB VRAM?
Yes, using the Q4_0 quantization which requires about 2.5 GB of VRAM at 8k context.
Do I need a special version of llama.cpp?
Yes, a fork that supports vision for Qwen 2.5 VL is required; the main branch does not support vision.
Is there a license restriction?
The license is not specified in the provided data; check the model card for details.