2 ms·
I did some minimal testing, mamba uses about 60% of VRAM in comparison to RetNet (parallel forward mode) with the model of the same size and the vocabulary of s
by pk-protect-ai 3y ago
I did some minimal testing, mamba uses about 60% of VRAM in comparison to RetNet (parallel forward mode) with the model of the same size and the vocabulary of same size during inference.