LI

LightOnOCR 3 4B

LightonocrGGUF by mradermacherUpdated 2026-10-09
Download on Hugging Face
Parameters4.21B
Context256K
Layers32
Architectureqwen35
Smallest quant1.9 GB
Quick answer

LightOnOCR 3 4B needs about 4.3 GB of memory at Q4_K_M (a 2.7 GB file) with an 8K context window, so it fits on a 6 GB GPU. This repository has 12 quants, from 1.9 GB (Q2_K) to 8.4 GB (F16). Longer context needs more: at 32K tokens the KV cache alone adds about 4.3 GB.

LightOnOCR 3 4B – lightweight OCR with vision and layout understanding

LightOnOCR 3 4B is a 4.21 billion‑parameter vision‑language model based on Qwen3.5. It transcribes text, outputs bounding boxes, and provides visual descriptions. The model is released under Apache 2.0 and supports transcription and grounding modes.

Good at

  • OCR transcription
  • Bounding‑box grounding
  • Visual description of images

Limits

  • No text‑only mode beyond transcription
  • Requires image input for vision features
  • Large VRAM needed for higher‑quality quants

Best for

  • Document OCR pipelines
  • Layout‑aware data extraction
  • Image‑to‑text conversion

Who it’s for. Researchers and developers needing a single model for OCR, layout analysis, and visual understanding. The model is suitable for both academic and commercial use under the Apache 2.0 license.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
Q2_K1.9 GB 3.5 GB
Fits Get file →
Q3_K_S2.1 GB 3.6 GB
Fits Get file →
Q3_K_M2.3 GB 3.8 GB
Fits Get file →
Q3_K_L2.4 GB 4.0 GB
Fits Get file →
IQ4_XS2.5 GB 4.1 GB
Fits Get file →
Q4_K_S2.6 GB 4.1 GB
Fits Get file →
Q4_K_M2.7 GB 4.3 GB
Fits Get file →
Q5_K_S3.0 GB 4.6 GB
Fits Get file →
Q5_K_M3.1 GB 4.6 GB
Fits Get file →
Q6_K3.5 GB 5.0 GB
Fits Get file →
Q8_04.5 GB 6.1 GB
Fits Get file →
F168.4 GB 10.0 GB
Too big Get file →

For 8 GB VRAM, use Q4_K_S (2.56 GB file, ~4.1 GB memory). For 12 GB VRAM, Q5_K_M (3.07 GB file, ~4.6 GB memory) offers better quality. If memory is limited, Q3_K_S (2.07 GB file, ~3.6 GB memory) is a good compromise.

What GPU do you need for LightOnOCR 3 4B?

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 Q3_K_L (2.4 GB file, about 4.0 GB in use).
  • 6 GB VRAM: the best quant that fits is Q6_K (3.5 GB file, about 5.0 GB in use).
  • 8 GB VRAM: the best quant that fits is Q8_0 (4.5 GB file, about 6.1 GB in use).
  • 12 GB VRAM: the best quant that fits is F16 (8.4 GB file, about 10.0 GB in use).

Which LightOnOCR 3 4B quant should you choose?

Q4_K_M is the usual starting point for LightOnOCR 3 4B: it balances file size and quality well. Here it is a 2.7 GB file.

The Q8_0 file is 4.5 GB, about 1.7 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.

If memory is tight, the smallest file is Q2_K at 1.9 GB, with the quality trade-offs that come with fewer bits. Use it only when memory is very tight. Expect a clear drop in quality, which is usually worse on small models and easier to live with on large ones.

New to quant names? Read GGUF quantization explained.

Specifications

Architecture from the GGUF header

Architectureqwen35
Parameters4.21B
Layers32
Trained context262,144 tokens
KV cache (fp16)128 KB per token

Source

Quantized bymradermacher
Licenseapache-2.0
Downloads (HF)3,119
Last updated2026-10-09

Frequently asked questions about LightOnOCR 3 4B

How much VRAM does LightOnOCR 3 4B need?

About 4.3 GB at Q4_K_M with an 8K context window. That is the 2.7 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 3 4B on an 8 GB GPU?

Yes, with Q8_0: the 4.5 GB file needs about 6.1 GB at 8K context. Larger quants in this repository do not fit in 8 GB.

What is the file size of LightOnOCR 3 4B GGUF?

It depends on the quant: Q2_K is 1.9 GB, Q3_K_S is 2.1 GB, Q3_K_M is 2.3 GB, Q3_K_L is 2.4 GB, IQ4_XS is 2.5 GB, Q4_K_S is 2.6 GB and 6 more.

What context length does LightOnOCR 3 4B support?

The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 1.1 GB at 8K and 4.3 GB at 32K.

Where can I download LightOnOCR 3 4B GGUF?

The files are on Hugging Face in the mradermacher/LightOnOCR-3-4B-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 3 4B 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 3 4B on 8 GB VRAM?

Yes, the Q4_K_S quant uses about 4.1 GB memory at 8 k context, fitting within 8 GB VRAM.

Does the model support text‑only transcription?

Yes, call it with an empty prompt and it returns full page text in markdown.

What license applies to this model?

The model is released under the Apache 2.0 license for research and commercial use.

Other GGUFs of LightOnOCR-3-4B

Similar models

Browse all →

More from mradermacher

All by mradermacher →

More Lightonocr models

All Lightonocr →

Other small models

All small models →