技術摘要(英)
(1) Diversified paired-data generation system: generates diversified paired data for supervised finetuning to improve RTL GenAI's code-generation performance. (2) Coarse-to-fine instruction tuning technique: boosts GenAI's understanding of brief prompts to enhance the accuracy of generated RTL code. (3) AutoML HPO: finds optimal hyperparameters for model deployment, making industrial GenAI application more accessible.