rontolisp:make-quantized-matrix

(rontolisp:make-quantized-matrix format dims)

Builds an all-zero quantized matrix (rontolisp:quantize) in format (q8-0) with the dimensions dims -- an integer for rank 1, or a list of one or two integers, the last a multiple of 32. Its purpose is to be the destination of a read-sequence: a quantized matrix's storage IS ggml's block layout, 34 bytes for every 32 elements, so a Q8_0 tensor in a GGUF file is one bulk transfer into it and write-sequence writes those same bytes back out. That is how gguf:read loads a Q8_0 tensor.

read-sequence and write-sequence count BYTES for this buffer (:start / :end too), 34 per block; a full transfer of a (rows cols) matrix moves rows * cols / 32 * 34 of them.

Interpreter and JVM only. On the WASM backends the call signals at run time, so a reader's Q8_0 arm compiles everywhere and refuses only when reached.