Jzz

torch round error

1

https://github.com/666DZY666/micronet/blob/096d9911b92aec9c52a5ddf1d8d694c32f345111/micronet/compression/quantization/wqaq/iao/quantize.py#L135 比如input = -2.5，  感觉需要添加一个判断 `tensor[tensor < 0] = torch.ceil(tensor[tensor < 0] - 0.5)`

Dose aimet Bias quant sim is the same with SNPE dsp quantize?

4

I noticed that，SNPE support bias quant bitwidth 8/32，In DSP，how the DSP accumulate the 8/32 quantized bias into the int32 accumulator ? how do aimet simulate the quant bias add ?

how tor check the register buffer params 'zero point '

1

i want to check the register buffer value 'zero_point' and 'scale' by print mode.state_dict, but model.state_dict seems do not record 'zero_point' and 'scale' ? how can i check the 'zero...

Mx FP Quantization About Subnorm

1

When quantizing Mx fp, the quantization scales of subnormal and normal values ​​should be different. Why does L394 clip to min_exp? I understand that it should clip to 1. Looking...