RUN THIS LLM
Search local LLM hardware requirements
DeepSeek OCR-2
DeepSeek · 3B MoE · Vision
3B MoE vision model for document OCR. Converts complex layouts to structured markdown using 144-256 visual tokens per page. 73 on OmniDocBench.
VRAM Requirements
| Quantization | VRAM |
|---|---|
| Q4_K_M (smallest) | 1.8 GB |
| Q8_0 (balanced) | 3.3 GB |
| FP16 (full quality) | 6 GB |
Specifications
- Parameters: 3B MoE (0.57B active per token)
- Category: Vision
- Max context: 8K tokens
- System RAM: 4 GB minimum
- HuggingFace: deepseek-ai/DeepSeek-OCR-2
Loading interactive analysis...