Qwen2.5-VL-7B-Instruct-GGUF

Qwen2.5-VL-7B-Instruct-GGUF

Unsloth's GGUF files of Qwen2.5-VL-7B-Instruct, a vision-language model that reads images, charts, documents and long videos and can output coordinates and structured data. For people who run it locally.

Made by unsloth on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: This repository holds Unsloth's GGUF files of Qwen2.5-VL-7B-Instruct, the instruction-tuned 7B vision-language model from the Qwen team. It reads images and video along with text. The Hugging Face record lists 7.6B parameters and about 3.8 GB at 4-bit.

Who it suits: Developers and analysts who want a local model that can read documents, charts and screens, and teams that need structured data pulled out of images.

Industry | Who uses it | How they use it | Example
Finance and commerce | Operations teams | Pull structured fields from invoices, forms and tables | Turn an invoice scan into structured data
Software | Developers of agents | Let a model read a screen and decide on actions | Direct a computer or phone task
Media analysis | Video analysts | Find the relevant moments in long videos | Locate the segment where an event happens

You can skip it if you only need text, since a plain language model is lighter.

Where to use it:

• llama.cpp and other tools that read GGUF files
• Python with the transformers library for the original weights
• Windows, macOS or Linux, wherever the file fits in memory

Use case | Who | What they get back
Document and chart reading | Analysts | Text answers about text, charts and layouts in an image
Object location | Developers | Bounding boxes or points, with stable JSON for coordinates
Long video understanding | Video teams | Answers about videos over an hour, with the relevant segments pinpointed

How to set it up: You need a GGUF-capable runtime with vision support and enough memory for about 3.8 GB at 4-bit. The part of the card used here does not include run steps.

• Python and pip, for the download tool
• A runtime that supports GGUF and this vision model
• Disk space for the file you choose

1. Download the GGUF files.
2. Load them in a runtime that supports Qwen2.5-VL.
3. Send an image and a question.

pip install -U huggingface_hub
hf download unsloth/Qwen2.5-VL-7B-Instruct-GGUF

How to use it: Send the model an image or a video together with a question. It can describe what it sees, read text and charts, locate objects, and return structured output for forms and tables.

Example: You send a photo of an invoice and ask for the items and totals as JSON. The model replies with structured fields read from the image. The card states this ability, but the part used here shows no sample output.

For newcomers and experts: If you are new, try a clear photo of a chart and ask a question about it. Start with a single image before you try video.

For experts, the card describes dynamic resolution and frame-rate sampling for video, with time-aligned position encoding. It also describes a vision encoder that uses window attention to run faster. The family has 3B, 7B and 72B sizes, and this one is the 7B instruct model.

Image, video or text input
Vision encoder with window attention
Qwen2.5 language model
Text or JSON output

Good to know:

Title: MathVista testmini score, as listed in the card
Qwen2.5-VL-7B: 68.2
MiniCPM-o 2.6: 60.6
InternVL2.5-8B: 58.3
Qwen2-VL-7B: 58.2
GPT-4o-mini: 52.4

• Results: the scores come from the model authors' own table
• Video: the card says it can follow videos over one hour
• Language: the record lists English
• License: Apache 2.0
• Cost to run: no API key or paid service named
• Data it reaches: downloads files from Hugging Face, nothing else is named
• Not for: the part of the card used here names no limits

On Hugging Face, 13 November 2026

• Task: image text to text
• Library: transformers
• Parameters: 7.6B
• Weights: about 3.8 GB at 4-bit (estimate)
• Formats: GGUF
• Downloads in the last 30 days: 285,067
• Likes: 253
• License: apache-2.0
• Languages: en
• Based on: Qwen/Qwen2.5-VL-7B-Instruct
• Fine-tunes: 1
• Tags: transformers, gguf, qwen2_5_vl, image-text-to-text, multimodal, unsloth, conversational

Links

• Model: https://huggingface.co/unsloth/Qwen2.5-VL-7B-Instruct-GGUF
• Files: https://huggingface.co/unsloth/Qwen2.5-VL-7B-Instruct-GGUF/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download unsloth/Qwen2.5-VL-7B-Instruct-GGUF

Then follow the usage in its model card. Nothing in this kit runs by itself.