Language models have shown promising performance on the task of translating natural language questions into SQL queries (Text-to-SQL). However, most of the state-of-the-art (SOTA) approaches rely on powerful yet closed-source large language models (LLMs), such as ChatGPT and GPT-4, which may have the limitations of unclear model architectures, data privacy risks, and expensive inference overheads. To address the limitations, we introduce CodeS, a series of pre-trained language models with parameters ranging from 1B to 15B, specifically designed for the text-to-SQL task. CodeS is a fully open-source language model, which achieves superior accuracy with much smaller parameter sizes. This paper studies the research challenges in building CodeS. To enhance the SQL generation abilities of CodeS, we adopt an incremental pre-training approach using a specifically curated SQL-centric corpus. Based on this, we address the challenges of schema linking and rapid domain adaptation through strategic prompt construction and a bi-directional data augmentation technique. We conduct comprehensive evaluations on multiple datasets, including the widely used Spider benchmark, the newly released BIRD benchmark, robustness-diagnostic benchmarks such as Spider-DK, Spider-Syn, Spider-Realistic, and Dr.Spider, as well as two real-world datasets created for financial and academic applications. The experimental results show that our CodeS achieves new SOTA accuracy and robustness on nearly all challenging text-to-SQL benchmarks.

Recognizing violence in videos is significant for the automatic identification and assessment of violence content to restrict the access to violence for specific audiences such as children. Existing methods focus on violence detection, which is only able to recognize whether there exists violence or not. Differently, this paper handles the problem of video violence rating, which provides a more granular classification of violence levels. However, there is no publicly available database for video violence rating since it asks for fine-grained violence level annotations. Therefore, this paper introduces a large-scale violence rating database, which will be publicly released. Furthermore, we propose a multimodal violence rating model. Different from existing models, our model makes use of the token-based interaction and contrastive learning techniques. The token-based interaction is able to strengthen the feature representations and make full use of multimodal features. The contrastive learning can improve the performance of the model. To evaluate our model, a wide range of experiments are conducted, and experiment results show that our model outperforms existing methods. We have made the VioShot database publicly available for downloading athttps://sites.google.com/site/xiangtaooo/.