what data type does GGML quantization cast to on inference?

[aviadingo]

Hello, I am interested to know the casting process during inference where an input signal is going through a quantized layer.
If I understand correctly we have the input which is x, I guess the type is float right?
and the layer itself is qunatized, let's say INT8.

does the layer weights get casted back to floats for computation or does the signal gets casted to int instead?

would love to know where I can find the answer in the code as well.

Thanks