Unlimited OCR من بايدو: حوّل ملفاتك المصوّرة إلى نص ببلاش، صفحة وحدة أو PDF كاملBaidu's Unlimited OCR: Turn Scanned Files into Text for Free, One Page or a Whole PDF

Key terms

OCR

Technology that reads the words in an image or scanned file and turns them into text you can copy and edit.

Unlimited OCR

An open source 3B model from Baidu, released June 2026 and built on DeepSeek OCR. It reads dozens of pages in one pass instead of splitting them page by page.

Hugging Face Space

A demo page on Hugging Face that runs the model on their servers, so you can try it in the browser with no install.

GPU

The graphics card that runs the model on your machine. The model page says it is tested on NVIDIA cards with CUDA.

The steps

1. 1

   Step 1

   Try it free on its demo page

   Open the demo on Hugging Face. Drop a page image or a PDF into INPUT and pick Long if the file has many pages.

   Press START, wait for the text in OCR OUTPUT, then COPY.

   huggingface.co

   

   Tip: The demo uploads your file to Hugging Face servers, so test with an ordinary file, not a private contract.

2. 2

   Step 2

   Check the model and the license

   The official model page lives under the baidu account on Hugging Face. It shows the MIT license, free even for commercial use, and a 3B model size.

   Always download from that official account, not from a modified copy you cannot trace.

   huggingface.co

   

3. 3

   Step 3

   Test it on an Arabic file before you rely on it

   The model page gives no accuracy figure for Arabic. Before you build on it, take a real Arabic page from your own files, run it through the model and compare it with the original.

   The easiest check: give Claude the image and the output text and ask it to count the mistakes.

   PromptCompare the OCR output with the original

   Copy

   I attached an image of an Arabic page, and below it the text an OCR model extracted from it.
   
   Compare the text with the image line by line and give me:
   1. The number of wrong or missing words
   2. Any numbers or dates that changed
   3. Your verdict: is it good enough for [file type: contracts / invoices / books] or not
   
   The extracted text:
   [paste the text here]

4. 4

   Step 4

   Check that your machine can run it

   To run it locally, the model page says it is tested on NVIDIA GPUs with Python 3.12 and CUDA 12.9.

   No NVIDIA card? Stay with the demo page from step one.

   

5. 5

   Step 5

   Install the libraries

   Open a terminal, create a fresh Python environment, then install the libraries at the versions the model page lists.

   

   PromptInstall commands

   Copy

   python3 -m venv .venv && source .venv/bin/activate
   pip install torch==2.10.0 torchvision==0.25.0 transformers==4.57.1
   pip install Pillow==12.1.1 matplotlib==3.10.8 einops==0.8.2 addict==2.4.0 easydict==1.13 pymupdf==1.27.2.2 psutil==7.2.2

6. 6

   Step 6

   Turn a whole PDF into text in one pass

   Save the code below as read_pdf.py and run it on your file. It turns the PDF pages into images and gives them to the model together instead of page by page.

   The result is saved in the folder you set in output_path.

   

   Promptread_pdf.py from the model page

   Copy

   import os, sys, tempfile, torch, fitz
   from transformers import AutoModel, AutoTokenizer
   
   name = 'baidu/Unlimited-OCR'
   tok = AutoTokenizer.from_pretrained(name, trust_remote_code=True)
   model = AutoModel.from_pretrained(name, trust_remote_code=True, use_safetensors=True, torch_dtype=torch.bfloat16).eval().cuda()
   
   def pdf_to_images(path, dpi=300):
       doc = fitz.open(path); d = tempfile.mkdtemp(); out = []
       for i, page in enumerate(doc):
           p = os.path.join(d, f'page_{i+1:04d}.png')
           page.get_pixmap(matrix=fitz.Matrix(dpi/72, dpi/72)).save(p); out.append(p)
       return out
   
   model.infer_multi(tok, prompt='<image>Multi page parsing.', image_files=pdf_to_images(sys.argv[1]),
       output_path='[your/output/dir]', image_size=1024, max_length=32768,
       no_repeat_ngram_size=35, ngram_window=1024, save_results=True)

   Tip: The code follows the example on the model page itself. Only change the output path.

7. 7

   Step 7

   Keep sensitive files on your machine

   The key difference: the demo page uploads your file to a server, while running it locally keeps the file on your machine.

   So contracts, IDs and bank statements run locally only. Ordinary files are fine on the demo.