Kelvin_Stop_Scaring_Party@tg says to 電訊台 apply pruning and quantization to the model and deploy to hardware for inference