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. > > …
  • suoko@feddit.it
    101·
    2 days ago

    2025 is going to be the year of deepseek and Gwen. Let’s see if it’s going to be the year of riscV too with its AI chips like the SiFive Intelligence X390