Can i directly infer the type for a stripped binary not in train or test set

[hackhaye]

According to the Readme, when the model is in the test phase, the input of the model is pkl, I would like to ask is there any way to predict the type of stripped binary directly and output it?

[hackhaye]

I compiled a binary with flag "-g -fcf-protection=none -fno-eliminate-unused-debug-types -frecord-gcc-switches -pipe -fno-lto -fno-inline-functions -fno-inline-small-functions -fno-inline-functions-called-once -fno-inline -O2",and use the command "./TYGR datagen ./binaryDATASET.pkl" , why i get the following results:

Source #functions: 0 Well Formed #functions: 0 Source #vars: 0 Well Formed #vars: 0

it seems that no var and function at all,but it does has vars and functions in binary

[ChangZhu1997]

Hey! The reason why it shows no source functions/variables is bec there is no dwarf info. Right now the logic is if there is no dwarf info matched skip that variable.

Dwarf info is used to match each nodes to the ground truth type. Since it is in training/testing phase we need dwarf to evaluate the performance(e.g. accuracy), the dwarf info is only used for evaluation and the model does not know that.

If you want to directly output predicted result for stripped bins I would suggest you to modify glow/datagen.py to create pkl files without dwarf info(ground truth). Then modify test module to only output predicted result.

[hackhaye]

This work is great, but I found that on my 32GB device, using "--parallel", it was extremely easy to get stuck on a 5MB program. Even if I change the smaller pool_size, for example, pool_size=3, it will freeze.

Are there any optimization measures