LI

LightOnOCR 2 1B

LightonocrGGUF by wangjazzUpdated 2026-01-23
Download on Hugging Face
Parameters596M
Context16K
Layers28
Architectureqwen3
Smallest quant0.4 GB
Quick answer

LightOnOCR 2 1B needs about 1.8 GB of memory at Q4_K_M (a 0.4 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 3 quants, from 0.4 GB (Q4_K_M) to 1.2 GB (F16).

LightOnOCR 2 1B – 1B‑parameter vision‑language OCR model

LightOnOCR 2 1B is a 1B‑parameter end‑to‑end vision‑language model for OCR. It converts documents, PDFs, scans and images into clean, naturally ordered text. The model is trained on a large, high‑quality corpus with strong French, arXiv, and scan coverage.

Good at

  • Fast OCR on single GPU
  • High accuracy on scanned documents
  • Multilingual text extraction

Limits

  • Requires vision encoder, not quantized
  • Context length limited to 16384 tokens
  • No vision support in llama.cpp quantized models

Best for

  • Document digitization
  • Receipt processing
  • Scientific paper OCR

Who it’s for. Developers needing efficient OCR on GPU‑limited hardware. Researchers working on document analysis and multilingual OCR tasks. Teams building pipelines that require fast, accurate text extraction from images.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
Q4_K_M0.4 GB 1.8 GB
Fits Get file →
Q8_00.6 GB 2.1 GB
Fits Get file →
F161.2 GB 2.6 GB
Fits Get file →

For 8 GB VRAM use Q4_K_M (0.4 GB file, ~1.8 GB memory at 8 k context). For 12 GB VRAM Q8_0 (0.64 GB file, ~2.1 GB memory). For 16 GB VRAM F16 (1.2 GB file, ~2.6 GB memory). Estimates are approximate.

What GPU do you need for LightOnOCR 2 1B?

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 F16 (1.2 GB file, about 2.6 GB in use).

Which LightOnOCR 2 1B quant should you choose?

Q4_K_M is the usual starting point for LightOnOCR 2 1B: it balances file size and quality well. Here it is a 0.4 GB file.

The Q8_0 file is 0.6 GB, about 1.6 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.

New to quant names? Read GGUF quantization explained.

Specifications

Architecture from the GGUF header

Architectureqwen3
Parameters596M
Layers28
Trained context16,384 tokens
KV cache (fp16)112 KB per token

Source

Quantized bywangjazz
Licenseapache-2.0
Downloads (HF)1,796
Last updated2026-01-23

Frequently asked questions about LightOnOCR 2 1B

How much VRAM does LightOnOCR 2 1B need?

About 1.8 GB at Q4_K_M with an 8K context window. That is the 0.4 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 LightOnOCR 2 1B on an 8 GB GPU?

Yes, with F16: the 1.2 GB file needs about 2.6 GB at 8K context. Larger quants in this repository do not fit in 8 GB.

What is the file size of LightOnOCR 2 1B GGUF?

It depends on the quant: Q4_K_M is 0.4 GB, Q8_0 is 0.6 GB, F16 is 1.2 GB.

What context length does LightOnOCR 2 1B support?

The GGUF header lists a trained context length of 16,384 tokens (16K). The KV cache grows with the context you use: about 0.9 GB at 8K and 1.9 GB at 16K.

Where can I download LightOnOCR 2 1B GGUF?

The files are on Hugging Face in the wangjazz/LightOnOCR-2-1B-gguf repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.

What license does LightOnOCR 2 1B use?

Hugging Face lists the license as apache-2.0. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.

Can I run LightOnOCR 2 1B on 8 GB VRAM?

Yes, using the Q4_K_M quantized file (0.4 GB) which requires about 1.8 GB of VRAM at 8 k context.

Does the model support multilingual OCR?

Yes, the model was trained with strong French, arXiv, and scan coverage and supports multilingual text extraction.

Is the vision encoder quantizable?

No, the vision encoder (mmproj) should not be quantized as it significantly impacts image understanding quality.

Other GGUFs of LightOnOCR-2-1B

Similar models

Browse all →

More Lightonocr models

All Lightonocr →

Other tiny models

All tiny models →