search

huawei-noah / vega

AutoML tools chain
http://www.noahlab.com.hk/opensource/vega/
Other
845 stars 176 forks source link

Can't find quantized weights #224

Closed L-ED closed 2 years ago

L-ED commented 2 years ago

After finish of quant-ea algorithm i checked {vega_dir}/tasks/{experiment_number}/output/fully_train and nas directories. I loaded weights and assumed that they would be in 8 or 4 bit format, but weights was in float format. Is your search using fake-quantization?

dawncc commented 2 years ago

real quantization in pytorch is not supported, we define a fake-quantization operator in quant-ea algorithm for simulation.