SolMover: Smart Contract Code Translation Based on Concepts
Abstract
Talk at Berkeley RDI Research Meeting on SolMover for smart contract translation.
Key points:
- The challenge of generating Move smart contracts with limited training data
- Knowledge graph of Move concepts for retrieval-augmented generation
- Iterative refinement with planning, coding, and debugging agents
- Evaluation with open-source LLMs showing accuracy improvements
Takeaway: ConMover/SolMover addresses the low-resource code generation challenge by teaching LLMs concepts rather than syntax, enabling translation to new verifiable languages.
