一句话：提出用LLM增强深度学习框架间转译，以自动化处理算子语义不一致并提升转换一致性。
深度学习转译器用于在不同深度学习框架之间进行源到源的模型转换，从而简化部署。然而，它们往往无法弥合算子层面的语义不一致，导致API不兼容，例如参数不兼容、算子缺失等问题，进而造成转换失败或性能下降。针对这一问题，论文提出借助大语言模型增强深度学习转译，目标是提升转译自动化程度并保持语义一致性。摘要未给出具体方法细节与实验评估结果。
Deep learning (DL) transpilers perform source-to-source model conversion between DL frameworks, simplifying deployment. However, they often fail to bridge operator semantic inconsistencies, resulting in API incompatibilities (e.g., incompatible parameters, missing operators), which cause conversion failures or performance drops
@inproceedings{paperbot9037,
title = {Enhancing Deep Learning Transpilation with LLMs: Automation and Semantic Consistency},
author = {Xingpei Li and Zhijie Jiang and Zhouyang Jia and Yuanliang Zhang and Zheng Si and Haifang Zhou and Shanshan Li},
booktitle = {FSE},
year = {2025},
doi = {10.1145/3696630.3731431}
}