There were over 75,000 machine learning models studied, with natural language processing models making up about 40% of tasks. NLP model categories like fill-in-the-blank and translation were most common. Model documentation has evolved from being model-centric to more data-centric, focusing on topics like fairness, bias, intended uses and limitations. Word embedding analysis showed terms related to evaluation and training changed between 2020 and 2022 to increasingly focus on responsible AI topics. While ML models are growing exponentially, NLP tasks are dominated by a few major categories with seasonal patterns.