For Flash there are some excellent Q2/Q4 hybrids. I know that model was QAT so it handles Q4 better but the meta on quantization seems to be shifting a little bit to be more intelligent about what exactly gets quantized.
For Flash there are some excellent Q2/Q4 hybrids. I know that model was QAT so it handles Q4 better but the meta on quantization seems to be shifting a little bit to be more intelligent about what exactly gets quantized.