Skip to content

Latest commit

 

History

History
71 lines (55 loc) · 1.98 KB

README.md

File metadata and controls

71 lines (55 loc) · 1.98 KB
Raw

🌴PalmTree

This is the official repository for 🌴PalmTree, which is a pre-train Language model for assembly. We will actively update it. Currently supported platforms: x86

You can find pre-trained PalmTree model here

please consider citing our paper Xuezixiang Li, Yu Qu, and Heng Yin, "PalmTree: Learning an Assembly Language Model for Instruction Embedding", [CCS'2021]

@inproceedings{li2021palmtree,
  title={Palmtree: learning an assembly language model for instruction embedding},
  author={Li, Xuezixiang and Qu, Yu and Yin, Heng},
  booktitle={Proceedings of the 2021 ACM SIGSAC Conference on Computer and Communications Security},
  pages={3236--3251},
  year={2021}
}

Acknowledgement:

This implementation is based on bert-pytorch, we add our training tasks for CFGs and DFGs.

Requirements:

  • cuda >= 10.1
  • pytorch >= 1.3.1
  • binary ninja (optional, for dataset generation)

Updates

extrinsic evaluations

We released code for extrinsic evaluations.

Extrinsic evluations including:

  • Gemini
  • EKALVYA

intrinsic evaluations

Intrinsic evlautions including:

  • Opcode Outlier detection
  • Operand Outlier detection
  • Basicblock matching

Dataset generator

dataflow_gen.py: generate dataflow dataset for PalmTree model via Binary Ninja and Binary ninja mid-level IR. control_flow_gen.py: generate control flow dataset for PalmTree model using Binary Ninja API.

Dataset format

TXT files. Data-flow and control-flow graph will be sampled by random walk algorithm. And then splitted into instruction pairs. For a given instruction sequence

push rbp
mov rbp, rsp
sub rsp, 0x20

Two lines of instruction pairs will be generated:

push rbp    mov rbp rsp
mov rbp rsp sub rsp 0x20

In detail:

push\<SPACE\>rbp<\t>mov\<SPACE\>rbp\<SPACE\>rsp<\n>

TODO:

  • Support more binary tools (Ghidra, IDA pro, etc.)
  • Support Docker