Take token count quantization of fused attention into consideration for CP results correction

Take token count quantization of fused attention into consideration for CP results correction #5929