Model Overview

raafatabualazm/decompiler-v2 is a specialized 4 billion parameter language model, fine-tuned from the Qwen/Qwen3-4B-Thinking-2507 base model. Developed by raafatabualazm, its primary function is idiomatic decompilation, translating assembly code into more readable, high-level programming languages.

Key Capabilities

• Assembly to High-Level Code Translation: Excels at converting assembly language into idiomatic high-level code, specifically demonstrated with Dart and Swift.
• Fine-tuned Architecture: Utilizes LoRA/DoRA adapters for efficient fine-tuning, leveraging custom datasets of assembly-to-Dart/Swift pairs.
• Base Model: Built upon the robust Qwen3-4B-Thinking-2507 architecture, providing a strong foundation for its specialized task.

Use Cases

This model is particularly well-suited for:

• Reverse Engineering: Assisting in the analysis and understanding of compiled binaries by generating human-readable source code.
• Code Porting: Facilitating the migration of legacy assembly codebases to modern high-level languages.
• Security Research: Aiding in vulnerability analysis by providing clearer insights into program logic from compiled executables.

Developers can load and utilize this merged model with standard Hugging Face transformers library calls, specifying raafatabualazm/decompiler-v2 as the repo_id.