MiMo-V2.6 Flash on 2X 6kPro?
#1
by mtcl - opened
I see this as TP4, weights look like they will fir on 2x 6Kpros, any chance this will work?
MiMo-V2.6 Flash with MXFP4 routed experts, block-FP8 projections, and
image, video, and audio encoders. Qualified at TP4 on SM120 with b12x
loading, dense linears, routed experts, and 192/128 language attention.
Requires the corresponding vLLM MiMo integration and block-FP8 output
padding at TP4.