There's a new PR for llamacpp claiming to boost prompt processing with rocm by around 15%, also fixes a bug which makes Q2_K 28x faster
There's a new PR for llamacpp claiming to boost prompt processing with rocm by around 15%, also fixes a bug which makes Q2_K 28x faster — reported by reddit.com, aggregated and ranked by ClawDigest.