llama.cpp

b9dfc25c - ggml : fix overflows in elu function (#8866)

ggml : fix overflows in elu function (#8866) It's helpful to use expm1f(x), because expf(x)-1 will result in overflow for 25% of single-precision floating point numbers.