RUN THIS LLM
Search local LLM hardware requirements
DeepSeek VL2 27B
DeepSeek · 27.5B MoE · Vision
27.5B MoE vision-language model with 4.5B active params and MLA attention. Flagship of the VL2 family for document, chart, and grounding tasks.
VRAM Requirements
| Quantization | VRAM |
|---|---|
| Q4_K_M (smallest) | 16.5 GB |
| Q8_0 (balanced) | 30.3 GB |
| FP16 (full quality) | 55 GB |
Specifications
- Parameters: 27.5B MoE (4.5B active per token)
- Category: Vision
- Max context: 4K tokens
- System RAM: 24 GB minimum
- HuggingFace: deepseek-ai/deepseek-vl2
Loading interactive analysis...