Update conversion script and include more triple weights to fp8

Add FP8 quantized model

initial commit