Comment by Marat_Dukhan

Yes, these optimizations work with existing tflite models, so long as the quantized operators they use are supported in XNNPACK.