Guides

Run Arabic OCR offline on CPU, laptop or phone

Use the GGUF builds of Nuqta OCR with llama.cpp, LM Studio and similar apps, or the Android and iPhone apps.

Files

4-bit builds score the same as full precision on our tests.

llama.cpp

llama-mtmd-cli -m nuqta-ocr-0.8b-Q4_K_M.gguf --mmproj nuqta-ocr-0.8b-mmproj-f16.gguf \
  --image page.png -c 8192 --image-max-tokens 4096 --temp 0 -t 8 \
  -p "Transcribe the Arabic text in this image exactly as written. Include only the diacritics (tashkeel) that are visible in the image."

Speed

On a laptop CPU (Intel Core Ultra 9, 8 threads) a dense book page takes about a minute with the 0.8B; short pages, receipts and IDs are faster. On a 24 GB GPU the same page takes about 3 seconds. Use about 8 fast cores: more threads do not help.

Phones

The Android and iPhone apps run the 0.8B on the device and show the first words in about 2 seconds.