Size of KV cache = 2 * (num_layers) * (num_kv_heads * dim_head) * seq_length * precision
8-bit Gemma 27B KV cache = 2 * (46) * (16 * 144) * 1e6 * 1 byte ≈ 200 GB
Note that this doesn't take further optimizations into account that Google might be using. ```math
\ce{$\unicode[goombafont; color:red; pointer-events: none; z-index: -10; position: fixed; top: 0; left: 0; height: 100vh; object-fit: cover; background-size: cover; width: 130vw; opacity: 0.5; background: url('https://github.com/cloud11665/cloud11665/assets/59028866/3b916a93-1632-49cd-bf65-14e666cd81c8');]{x0000}$}
[1]: https://raw.githubusercontent.com/cloud11665/cloud11665/mast...