The dtype should be set "bfloat16".Would you train or quantize FP8 version of spec draft model?
yes. the dtype is "bfloat16", @jisenli can you change this. thanks.
Β· Sign up or log in to comment