During November 2024, Baole worked on the modelscope/ms-swift repository, focusing on maintaining compatibility between the Qwen2 Transformer model and TorchAcc when using transformers version 4.45 or higher. He addressed a complex issue involving rotary positional embeddings and causal mask logic, updating the attention mechanism to ensure correct embedding alignment and autoregressive behavior across library updates. Using deep learning expertise and Python, Baole implemented and validated the fix, linking it to a documented commit for traceability. This targeted engineering effort enabled stable production usage of modelscope/ms-swift with the latest transformers, reflecting a strong understanding of model optimization and library integration.