* gguf_convert_endian.py: skip MXFP4 data * Use gguf.constants.GGML_QUANT_SIZES to determine block sizes