Conversion older ggml code to latest version

[HoiV]

Hi, to learn more from GGML, I am trying to convert some older apps to run with the latest GGML lib.

In doing so, I ran across this type of construct:
inpL = ggml_mul_mat(ctx0, inpL, sum);

    // normalizer
    struct ggml_tensor *sqr1 = ggml_sqr(ctx0, inpL);
    struct ggml_tensor *sum1 = ggml_sum(ctx0, sqr1);
    struct ggml_tensor *sqrt1 = ggml_sqrt(ctx0, sum1);
    inpL = ggml_scale(ctx0, inpL, ggml_div(ctx0, ggml_new_f32(ctx0, 1.0f), sqrt1));

Since ggml_scale() only accepts a float for the 3rd parameter, what is the best way to get the right value from:

"ggml_div(ctx0, ggml_new_f32(ctx0, 1.0f), sqrt1)"?

Thank you in advance.

[slaren]

ggml_mul now supports broadcasting in all the dimensions, so you can use it in place of ggml_scale.