Large Language Models (LLMs) are prone to generating content that exhibits gender biases, raising significant ethical concerns. Alignment, the process of fine-tuning LLMs to better align with desired behaviors, is recognized as an effective approach to mitigate gender biases. Although proprietary LLMs have made significant strides in mitigating gender bias, their alignment datasets are not publicly available. The commonly used and publicly available alignment dataset, HH-RLHF, still exhibits gender bias to some extent. There is a lack of publicly available alignment datasets specifically designed to address gender bias. Hence, we developed a new dataset named GenderAlign, aiming at mitigating a comprehensive set of gender biases in LLMs. This dataset comprises 8k single-turn dialogues, each paired with a “chosen” and a “rejected” response. Compared to the “rejected” responses, the “chosen” responses demonstrate lower levels of gender bias and higher quality. Furthermore, we categorized the gender biases in the “rejected” responses of GenderAlign into 4 principal categories. The experimental results show the effectiveness of GenderAlign in reducing gender bias in LLMs.

The bigger the corpus, the more topics it can potentially support. To truly make full use of massive text corpora, a topic model inference algorithm must therefore scale efficiently in 1) documents and 2) topics, while 3) achieving accurate inference. Previous methods have achieved two out of three of these criteria simultaneously, but never all three at once. In this paper, we develop an online inference algorithm for topic models which leverages stochasticity to scale well in the number of documents, sparsity to scale well in the number of topics, and which operates in the collapsed representation of the topic model for improved accuracy and run-time performance. We use a Monte Carlo inner loop in the online setting to approximate the collapsed variational Bayes updates in a sparse and efficient way, which we accomplish via the MetropolisHastings Walker method. We showcase our algorithm on LDA and the recently proposed mixed membership skip-gram topic model. Our method requires only amortized O(k_d) computation per word token instead of O(K) operations, where the number of topics occurring for a particular document k_d≪ the total number of topics in the corpus K, to converge to a high-quality solution.