Dear experts,
Our customer found that there are 4 kinds of quantization method(TIDL_QuantStyleFixed, TIDL_QuantStyleDynamic, TIDL_QuantStyleNP2Fixed, TIDL_QuantStyleP2Dynamic). import tool help says:
{(char *)"quantizationStyle", &gParams.quantizationStyle , INT_32, SINGLE, false, false, NULL,
"Quantization method: see --help-quantizationStyle",
"Quantization method used for quantizing parameters. The following types are currently supported\n"
" 0 : Caffe-jacinto specific\n"
" 1 : Dynamic\n"
" 2 : Fixed\n"
" 3 : Power-of-2 Dynamic\n"
"\n"
"This parameter accepts an integer (0, 1, 2 or 3) indicating which\n"
"quantization type to be used\n"
"\n"
"The default value is \"2\" (Fixed)\n"},
In our docs, it only says we support NP2 & P2.
Do we support TIDL_QuantStyleDynamic? What if we set the quant method to TIDL_QuantStyleDynamic?
Thanks & Best Regards!
ZM