Xiaomi-OCR-0-GGUF

Xiaomi-OCR-0-GGUF

Provides GGUF files of Xiaomi-OCR-0, a compact 0.8B vision-language model for reading documents and text in images, in sizes from about 466 MB to 1.52 GB. For local OCR with llama.cpp.

Made by prithivMLmods 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 GGUF files of Xiaomi-OCR-0, a small vision-language model built for OCR and document parsing. It has about 0.8B parameters and is based on Qwen3.5-0.8B-Base.

The card says it was trained on a corpus of roughly 170M OCR-focused samples. The files come in several quantization sizes, plus a projection file that lets the model read images.

Who it suits: It suits people who want to turn scans, screenshots and documents into text on their own computer, with a model small enough for modest hardware.

Industry | Who uses it | How they use it | Example
Office work | Administrators | Pull text out of scanned pages | Read a scanned form
Software | Developers | Add local OCR to an app | Parse a screenshot into text
Finance | Back-office teams | Read printed receipts and statements | Capture lines from a receipt
Research | Document AI engineers | Test a compact OCR model | Compare quant sizes

If you need an OCR service with a hosted API or support for many languages, you can skip it, since the card lists English only.

Where to use it: The files run in llama.cpp compatible tools on Windows, macOS or Linux, and the card names llama.cpp as the engine used for the quantization.

• llama.cpp and tools built on it
• Small laptops, since the files start at about 466 MB
• Pictures plus a text prompt as input

Use case | Who | What they get back
Document parsing | Developers | Text read from page images
Screenshot reading | Everyday users | Text taken from a screen capture
Size testing | Engineers | A comparison across the quant files

How to set it up: You need Python with pip to fetch the files and a tool that loads GGUF files. Pick one quant file, and add the projection file so the model can read images.

• Python with pip
• llama.cpp or a similar tool
• Less than 2 GB of disk space for any file here

1. Install the Hugging Face hub package.
2. Download the repository, or just the quant and the projection file.
3. Load the quant file in your GGUF tool with the projection file.

pip install -U huggingface_hub
hf download prithivMLmods/Xiaomi-OCR-0-GGUF

How to use it: Load the quant file and the projection file in your tool, give it an image of a document and a short instruction, and read the text it returns. The card gives no run command, so follow your tool's instructions.

File | Size | When to pick it
BF16 | 1.52 GB | You want full weights and the best quality
Q6_K | 630 MB | You want near-full quality
Q5_K_M | 578 MB | You want high quality
Q4_K_M | 529 MB | The balanced default for most uses
Q4_K_S | 505 MB | You want to save a little more space
Q3_K_L | 491 MB | Memory is low and lower quality is fine
Q3_K_M | 466 MB | The smallest file, with lower quality
mmproj-bf16 | 207 MB | You need the image projection

Example: You load the Q4_K_M file with the projection file and give it a photo of a printed page. You ask it to write out the text, and it answers with the page's words as plain text.

For newcomers and experts: If you are new, start with the Q4_K_M file, one clear page photo and a short request to read the text.

If you know the field, the original model combines a text anchoring method called Q-Mask with extra pretraining and reinforcement learning across several tasks. Quantization then trades some accuracy for a smaller file.

Pick a quant file
Load it with the projection file
Give it a page image
Read the extracted text

Good to know: The smaller quants lose some accuracy, so test a few on your own documents. Check that any private pages stay on your own machine, since the model runs locally.

• License: apache-2.0
• Cost to run: none named
• Data it reaches: files you download, and the images you give it
• Not for: languages other than English, since the card lists English only

On Hugging Face, 13 November 2026

• Task: image text to text
• Library: transformers
• Parameters: 752M
• Weights: about 376 MB at 4-bit (estimate)
• Formats: GGUF
• Downloads in the last 30 days: 44,340
• Likes: 19
• License: apache-2.0
• Languages: en
• Based on: xiaomi-research/Xiaomi-OCR-0
• Tags: transformers, gguf, llama-cpp, document-parsing, document-understanding, visual-question-answering, key-information-extraction, vision-language-model, multimodal, xiaomi, seerray, image-text-to-text

Links

• Model: https://huggingface.co/prithivMLmods/Xiaomi-OCR-0-GGUF
• Files: https://huggingface.co/prithivMLmods/Xiaomi-OCR-0-GGUF/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download prithivMLmods/Xiaomi-OCR-0-GGUF

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