deepseek-ocr2 needs about 4.5 GB of memory at Q8_0 (a 3.6 GB file) with an 8K context window, so it fits on a 6 GB GPU. This repository has 5 quants, from 2.3 GB (Q4_K) to 6.8 GB (F16).
DeepSeek-OCR2 CrispEmbed GGUF model for OCR embedding
DeepSeek-OCR2 is a 3.39B parameter model converted to GGUF for use with CrispEmbed. It uses a SAM-ViT-B encoder, Qwen2 encoder, and a DeepSeek-V2 MoE decoder. The model is licensed under apache-2.0.
Good at
- OCR embedding
- SAM patch embedding
- MoE decoding
Limits
- No vision support beyond OCR
- Requires 8k context for estimates
- License is apache-2.0
Best for
- OCR embedding pipelines
- Text extraction from images
- Embedding generation for document search
Who it’s for. Developers building OCR or document processing systems that need efficient embeddings. The model is suitable for applications requiring a lightweight, quantized version of DeepSeek-OCR2.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q4_K | 2.3 GB | 3.1 GB | Fits | Get file → | |
| Q4_K | 2.3 GB | 3.1 GB | Fits | Get file → | |
| Q8_0 | 3.6 GB | 4.5 GB | Fits | Get file → | |
| F16 | 6.8 GB | 8.1 GB | Too big | Get file → | |
| F16 | 6.8 GB | 8.1 GB | Too big | Get file → |
For 8 GB VRAM, use Q4_K (2.31 GB, ~3.1 GB memory). For 12 GB VRAM, use Q8_0 (3.61 GB, ~4.5 GB memory). For larger memory budgets, F16 (6.78 GB, ~8.1 GB memory) provides higher precision.
What GPU do you need for deepseek-ocr2?
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 Q4_K (2.3 GB file, about 3.1 GB in use).
- 6 GB VRAM: the best quant that fits is Q8_0 (3.6 GB file, about 4.5 GB in use).
- 8 GB VRAM: the best quant that fits is Q8_0 (3.6 GB file, about 4.5 GB in use).
- 12 GB VRAM: the best quant that fits is F16 (6.8 GB file, about 8.1 GB in use).
Which deepseek-ocr2 quant should you choose?
Q8_0 (3.6 GB) is the middle option in this repository. Close to lossless. It costs roughly twice the memory of a 4-bit file, so it suits cases where you have plenty of VRAM.
If memory is tight, the smallest file is Q4_K at 2.3 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 deepseek-ocr2
How much VRAM does deepseek-ocr2 need?
About 4.5 GB at Q8_0 with an 8K context window. That is the 3.6 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-ocr2 on an 8 GB GPU?
Yes, with Q8_0: the 3.6 GB file needs about 4.5 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of deepseek-ocr2 GGUF?
It depends on the quant: Q4_K is 2.3 GB, Q4_K is 2.3 GB, Q8_0 is 3.6 GB, F16 is 6.8 GB, F16 is 6.8 GB.
Where can I download deepseek-ocr2 GGUF?
The files are on Hugging Face in the cstr/deepseek-ocr2-crispembed-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 deepseek-ocr2 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 DeepSeek-OCR2 on 8 GB VRAM?
Yes, using the Q4_K quantized file (2.31 GB) which requires approximately 3.1 GB memory at 8k context.
Does this model support vision tasks beyond OCR?
No, it is specifically designed for OCR embedding and does not provide general vision capabilities.
What license applies to this GGUF model?
The model is distributed under the apache-2.0 license, matching the upstream license.