pub fn quantize_head(q: &[f32], words: Words, scaling: ActScaling) -> ActRowExpand description
A query head as the array reads it: per 32 elements (the head padded with zeros to a whole group), or the whole head one block.
pub fn quantize_head(q: &[f32], words: Words, scaling: ActScaling) -> ActRowA query head as the array reads it: per 32 elements (the head padded with zeros to a whole group), or the whole head one block.