ggml : x2 speed for WASM by optimizing SIMD
simonwillison.net
PR by Xuan-Son Nguyen for `llama.cpp`: > This PR provides a big jump in speed for WASM by leveraging SIMD instructions for `qX_K_q8_K` and `qX_0_q8_0` dot product functions. > > …
  • Tachanka [comrade/them]@hexbear.netEnglish
    22·
    3 days ago

    It is 2025

    AI writes code to update itself

    I still have to load the dishwasher by hand

    I still have to change the baby’s diapers

    I still have to go to work tomorrow

    Do things ever happen?

    Some say nothing ever happens.

    Others argue that everything always happens.

    I love the real movement.