* fix(l4): fix fp8 logic on l4 * also quant weights with single scale * use marlin even on 89
Attach a file by drag & drop or click to upload
