← All models

DeepSeek-VL2-Tiny

DeepSeek · 2024-12-13 · 3.4B parameters

DeepSeek-VL2-Tiny is the smallest model of DeepSeek-VL2, released December 13, 2024: a SigLIP vision encoder with dynamic tiling feeding a DeepSeekMoE language model (12 layers, multi-head attention, 64 routed and 2 shared experts, 6 routed active). Its released weights contain 3,370,501,440 parameters, about 1.0B activated per token by DeepSeek's rounded figure, with a 4,096-token context.

Benchmark scores