Abstract
Heterogeneous database or code migration entails the conversion of schema and/or code from a source language to a target language. Current tools only partially address the database migration problem and full migration requires labor and time intensive manual developer intervention. This disclosure describes techniques to reduce or eliminate the human intervention required for code/schema migration by automatically identifying and categorizing recurring instances of objects or code that are resistant to automatic translation, referred to as gaps in translation. A large language model (LLM) is trained, e.g., using few-shot learning, to translate such gaps in translation using developer-provided training examples illustrating the translation of the object/code segments from different types of gaps from the source to the target language. Once trained, the LLM can be utilized to translate the rest of the code base.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
n/a, "Compiler Categorization and LLM-based Code Translation via Few-shot learning", Technical Disclosure Commons, (March 08, 2024)
https://www.tdcommons.org/dpubs_series/6766