A neural-based decompiler using a Transformer model for decompilation tasks.

• The model only supports mips assembly decompilation for now.
• The model has been trained on examples from math.h so it is not usable for now with other corpus of code.

In this project, we want to demonstrate if Deep Learning and especially Transformer models can be applied to decompilation tasks. Transformer models are used for common NLP tasks and is the state of the art in this field. We think that decompilation can be seen as a translation task, where we want to translate a low level programming language (PL) to a higher level programming language. For now, the model uses as low level PL the MIPS Assembly and as high level PL the C language.

python3 decompile.py --train --model-path path_to_save

python3 decompile.py --decompile path/to/asm-file.s --model-path path_to_save

The architecture used in this project is the same transformer used in the orginial paper All you need is attention with some improvements and adaptatations to be applied to the decompilation task.