8 ms·deepseek-v2,v3,r1 are all using multi-headed attention.by MichaelMoser123 1y agodeepseek-v2,v3,r1 are all using multi-headed attention.