My focus are models that handle multimodal data (language, vision, audio) and how to speed up inference with tools such as sparsity, distillation, pruning, model cascades, early exits, or quantization.