Jerry Li

Machine Learning Engineer Intern @Nexa Ai

CA, US

Conducting R&D on multimodal RAG methods Automated the conversion of PyTorch models to CoreML for streamlined deployment workflows for Apple devices. Accelerated Whisper inference on Apple Neural Engine via fixed-length KV caching, MHA splitting, and memory-access tuning in PyTorch, resulting a 2x speedup compared to GPU execution. Parallelized the Encoder and Decoder of Whisper with multi-threading inference across both GPU and ANE for real-time audio transcription