Waadec 27, 202395 views

WaDec is an approach leveraging a fine-tuned LLM to decompile Wasm binary code into a more comprehensible source code. The model's training on a specialized wat-c dataset, coupled with self-supervised learning, has proven pivotal in enhancing decompilation efficacy. Our results indicate that WaDec significantly outperforms existing tools, achieving a minimal code inflation rate and maintaining high recompilability and re-execution rates.

This advancement not only bolsters the readability and analyzability of Wasm code but also paves the way for more robust automated code analysis, optimization, and security auditing processes. Our dataset is specifically designed for decompiling WebAssembly Wasm. The main features of the dataset are as follows:. For infering, please run infering.

Waadec 27, 202395 views: USTR invites the public

In Section 5. To compute these metrics, please run eval. Skip to content. You signed in with another tab or window.