search

neurosim / DNN_NeuroSim_V1.3

Benchmark framework of compute-in-memory based accelerators for deep neural network (inference engine focused)
62 stars 36 forks source link

VGG 使用4-bit ADC的error #4

Closed zhangfeixiang222 closed 3 years ago

zhangfeixiang222 commented 3 years ago

您好，我在使用你们提供的vgg8.pth，使用inference=1，设置ADC的精度为4-bit，输出结果很糟糕；如果在训练时将inference=1，会导致训练速度很慢。请问在manual中显示4-bit ADC时仍能达到90%左右的准确率，请问是在什么条件下测出来的呢? 谢谢！

neurosim commented 3 years ago

Nonlinear quantization ADC is used to get good accuracy with only 4-bit. You have to manually set the quantization thresholds for the nonlinear ADC because they are different under different conditions.

jimmyli-421 commented 2 years ago

@zhangfeixiang222 哈囉~我也遇到相同的問題，請問您有解決了嗎?