Add weight-only int8 quant of s3
#1
by sandpies - opened
Plain weight-only int8 (per-row scales, no activation calibration, no learned rounding) of s3.safetensors, made with convert_to_quant 1.3.4 (--int8 --scaling_mode row --comfy_quant --simple --minimaxh3). Loads as int8_tensorwise in ComfyUI. Size 20,948,314,532 bytes. sha256 c59bf4ddbcb062265ba0310397c57905e71cb9a893278d19007c3d7665514973
FusionCow changed pull request status to merged