DeepSeek OCR needs about 2.7 GB of memory at Q4_K_M (a 2.0 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 3 quants, from 2.0 GB (Q4_K_M) to 5.9 GB (BF16).
DeepSeek OCR – 2.93B parameter optical‑text compression model
DeepSeek OCR is a 2.93 billion‑parameter model designed for optical‑text compression. It is available in GGUF format for llama.cpp PR #17400. The model supports inference on NVIDIA GPUs with CUDA 11.8 and can be accelerated with vLLM.
Good at
- Optical‑text compression
- Efficient inference with low‑precision quants
- Compatible with llama.cpp MTMD branch
Limits
- No vision support in standard llama.cpp
- Requires PR branch of llama.cpp
- License not specified
Best for
- Compressing scanned documents
- Speeding up OCR pipelines
- Running on GPUs with limited VRAM
Who it’s for. Developers building OCR or document‑compression systems who need a lightweight, GPU‑friendly model. It is suitable for projects that can use the llama.cpp MTMD branch and have access to NVIDIA GPUs.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q4_K_M | 2.0 GB | 2.7 GB | Fits | Get file → | |
| Q8_0 | 3.1 GB | 4.0 GB | Fits | Get file → | |
| BF16 | 5.9 GB | 7.1 GB | Fits | Get file → |
For 8 GB VRAM, use Q4_K_M (1.95 GB file, ~2.7 GB memory at 8k context). For 16 GB VRAM, Q8_0 (3.13 GB file, ~4 GB memory) is a good balance. If more memory is available, BF16 (5.88 GB file, ~7.1 GB memory) offers higher precision.
What GPU do you need for DeepSeek OCR?
These figures add the file size, an fp16 KV cache at an 8K context window 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 Q8_0 (3.1 GB file, about 4.0 GB in use).
- 6 GB VRAM: the best quant that fits is Q8_0 (3.1 GB file, about 4.0 GB in use).
- 8 GB VRAM: the best quant that fits is BF16 (5.9 GB file, about 7.1 GB in use).
Which DeepSeek OCR quant should you choose?
Q4_K_M is the usual starting point for DeepSeek OCR: it balances file size and quality well. Here it is a 2.0 GB file.
The Q8_0 file is 3.1 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
Source
Frequently asked questions about DeepSeek OCR
How much VRAM does DeepSeek OCR need?
About 2.7 GB at Q4_K_M with an 8K context window. That is the 2.0 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 DeepSeek OCR on an 8 GB GPU?
Yes, with BF16: the 5.9 GB file needs about 7.1 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of DeepSeek OCR GGUF?
It depends on the quant: Q4_K_M is 2.0 GB, Q8_0 is 3.1 GB, BF16 is 5.9 GB.
Where can I download DeepSeek OCR GGUF?
The files are on Hugging Face in the sabafallah/DeepSeek-OCR-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
Can I run DeepSeek OCR on 8 GB VRAM?
Yes, the Q4_K_M quant uses 1.95 GB file size and requires about 2.7 GB memory at 8k context, fitting within 8 GB VRAM.
Does this model support vision tasks?
No, the model is not designed for vision tasks in standard llama.cpp; it requires the MTMD branch.
What is the recommended quant for 16 GB VRAM?
Use Q8_0, which is 3.13 GB in size and needs about 4 GB memory at 8k context.