Google's new Task-level Mixture-of-Experts method, a containerized inference solution for transformers, post-training quantization with TensorFlow Lite, and more