Please add INT8 ConvRot for text-encoder models like Qwen3-VL-4B/8B
#4
by twodog - opened
As your team stated: "FP8 is unnecessary for some models since INT8 ConvRot provides better quality and therefore also allows more layers to be quantized, ultimately making it faster on all NVIDIA GPUs."
twodog changed discussion title from Will text-encoder models such as Qwen3-VL-4B/8B all have INT8 ConvRot versions to replace FP8๏ผ to Please add INT8 ConvRot for text-encoder models like Qwen3-VL-4B/8B