Unlimited OCR 2 needs about 3.7 GB of memory at Q6_K (a 2.7 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 5 quants, from 2.0 GB (Q4_K) to 5.9 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 2.0 GB.
Unlimited OCR 2 – 3.3B Mixture‑of‑Experts vision model for unlimited‑horizon document parsing
Unlimited OCR 2 is a 2.94B parameter mixture‑of‑experts model trained on 131072 context tokens. It uses Sparse Reference Sliding Window Attention to parse documents of unlimited length while keeping memory usage stable. The model is vision‑enabled and requires a dedicated vision encoder.
Good at
- Long‑document OCR
- Layout‑aware text extraction
- Vision‑enabled parsing
Limits
- Requires llama.cpp with unlimited‑ocr2 support
- No text‑only inference without vision encoder
- Large VRAM needed for higher quant
Best for
- Scanning PDFs to structured text
- Automated form extraction
- Document digitization pipelines
Who it’s for. Developers building offline OCR solutions and researchers needing ultra‑long document parsing. The model is suited for environments where local, privacy‑preserving inference is required.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q4_K | 2.0 GB | 3.0 GB | Fits | Get file → | |
| Q5_K | 2.3 GB | 3.3 GB | Fits | Get file → | |
| Q6_K | 2.7 GB | 3.7 GB | Fits | Get file → | |
| Q8_0 | 3.1 GB | 4.1 GB | Fits | Get file → | |
| BF16 | 5.9 GB | 6.9 GB | Fits | Get file → |
For 8 GB VRAM, use Q4_K (file size 2.02 GB, ~3 GB memory at 8 k context). For 12 GB VRAM, Q5_K (2.28 GB, ~3.3 GB memory) is suitable. For 16 GB VRAM, Q6_K (2.74 GB, ~3.7 GB memory) or Q8_0 (3.13 GB, ~4.1 GB memory) can be used. Memory estimates are approximate.
What GPU do you need for Unlimited OCR 2?
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 Q6_K (2.7 GB file, about 3.7 GB in use).
- 6 GB VRAM: the best quant that fits is Q8_0 (3.1 GB file, about 4.1 GB in use).
- 8 GB VRAM: the best quant that fits is BF16 (5.9 GB file, about 6.9 GB in use).
Which Unlimited OCR 2 quant should you choose?
Q6_K (2.7 GB) is the middle option in this repository. Very close to the original quality for most uses, at a clearly larger size than 4-bit.
If memory is tight, the smallest file is Q4_K at 2.0 GB, with the quality trade-offs that come with fewer bits. The usual sweet spot between file size and quality, and the quant most people start with.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Unlimited OCR 2
How much VRAM does Unlimited OCR 2 need?
About 3.7 GB at Q6_K 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 Unlimited OCR 2 on an 8 GB GPU?
Yes, with BF16: the 5.9 GB file needs about 6.9 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Unlimited OCR 2 GGUF?
It depends on the quant: Q4_K is 2.0 GB, Q5_K is 2.3 GB, Q6_K is 2.7 GB, Q8_0 is 3.1 GB, BF16 is 5.9 GB.
What context length does Unlimited OCR 2 support?
The GGUF header lists a trained context length of 131,072 tokens (128K). The KV cache grows with the context you use: about 0.5 GB at 8K and 2.0 GB at 32K.
Where can I download Unlimited OCR 2 GGUF?
The files are on Hugging Face in the AtomicChat/Unlimited-OCR-2-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 Unlimited OCR 2 use?
Hugging Face lists the license as mit. 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 Unlimited OCR 2 on 8 GB VRAM?
Yes, the Q4_K quant uses about 3 GB of VRAM at 8 k context, which fits within an 8 GB budget.
Does it support text‑only inference?
No, the model requires the vision encoder to process images; text‑only inference is not supported.
What llama.cpp version is needed?
Build llama.cpp commit b11913 with the unlimited‑ocr2 patch from RX5950XT/Unlimited-OCR-2-GGUF.