Writing commit messages is a tedious daily task for many software developers, and often remains neglected. Automating this task has the potential to save time while ensuring that messages are informative. A high-quality dataset and an objective benchmark are vital preconditions for solid research and evaluation towards this goal. We show that existing datasets exhibit various problems, such as the quality of the commit selection, small sample sizes, duplicates, privacy issues, and missing licenses for redistribution. This can lead to unusable models and skewed evaluations, where inferior models achieve higher evaluation scores due to biases in the data. We compile a new large-scale dataset, CommitBench, adopting best practices for dataset creation. We sample commits from diverse projects with licenses that permit redistribution and apply our filtering and dataset enhancements to improve the quality of generated commit messages. We use CommitBench to compare existing models and show that other approaches are outperformed by a Transformer model pretrained on source code. We hope to accelerate future research by publishing the source code( https://github.com/Maxscha/commitbench ).
The field of deep generative modeling has grown rapidly and consistently over the years. With the availability of massive amounts of training data coupled with advances in scalable unsupervised learning paradigms, recent large-scale generative models show tremendous promise in synthesizing high-resolution images and text, as well as structured data such as videos and molecules. However, we argue that current large-scale generative AI models do not sufficiently address several fundamental issues that hinder their widespread adoption across domains. In this work, we aim to identify key unresolved challenges in modern generative AI paradigms that should be tackled to further enhance their capabilities, versatility, and reliability. By identifying these challenges, we aim to provide researchers with valuable insights for exploring fruitful research directions, thereby fostering the development of more robust and accessible generative AI solutions.
While (large) language models have significantly improved over the last years, they still struggle to sensibly process long sequences found, e.g., in books, due to the quadratic scaling of the underlying attention mechanism. To address this, we propose NextLevelBERT, a Masked Language Model operating not on tokens, but on higher-level semantic representations in the form of text embeddings. We pretrain NextLevelBERT to predict the vector representation of entire masked text chunks and evaluate the effectiveness of the resulting document vectors on three task types: 1) Semantic Textual Similarity via zero-shot document embeddings, 2) Long document classification, 3) Multiple-choice question answering. We find that next level Masked Language Modeling is an effective technique to tackle long-document use cases and can outperform much larger embedding models as long as the required level of detail is not too high. We make model and code available.
The emergence of various medical large language models (LLMs) in the medical domain has highlighted the need for unified evaluation standards, as manual evaluation of LLMs proves to be time-consuming and labor-intensive. To address this issue, we introduce MedBench, a comprehensive benchmark for the Chinese medical domain, comprising 40,041 questions sourced from authentic examination exercises and medical reports of diverse branches of medicine. In particular, this benchmark is composed of four key components: the Chinese Medical Licensing Examination, the Resident Standardization Training Examination, the Doctor In-Charge Qualification Examination, and real-world clinic cases encompassing examinations, diagnoses, and treatments. MedBench replicates the educational progression and clinical practice experiences of doctors in Mainland China, thereby establishing itself as a credible benchmark for assessing the mastery of knowledge and reasoning abilities in medical language learning models. We perform extensive experiments and conduct an in-depth analysis from diverse perspectives, which culminate in the following findings: (1) Chinese medical LLMs underperform on this benchmark, highlighting the need for significant advances in clinical knowledge and diagnostic precision. (2) Several general-domain LLMs surprisingly possess considerable medical knowledge. These findings elucidate both the capabilities and limitations of LLMs within the context of MedBench, with the ultimate goal of aiding the medical research community.
Efficiently training large language models requires parallelizing across hundreds of hardware accelerators and invoking various compute and memory optimizations. When combined, many of these strategies have complex interactions regarding the final training efficiency. Prior work tackling this problem did not have access to the latest set of optimizations, such as FlashAttention or sequence parallelism. In this work, we conduct a comprehensive ablation study of possible training configurations for large language models. We distill this large study into several key recommendations for the most efficient training. For instance, we find that using a micro-batch size of 1 usually enables the most efficient training layouts. Larger micro-batch sizes necessitate activation checkpointing or higher degrees of model parallelism and also lead to larger pipeline bubbles. Our most efficient configurations enable us to achieve state-of-the-art training efficiency results over a range of model sizes, most notably a Model FLOPs utilization of 70.5% when training a 13B model.
Intent detection and identification from multi-turn dialogue has become a widely explored technique in conversational agents, for example, voice assistants and intelligent customer services. The conventional approaches typically cast the intent mining process as a classification task. Although neural classifiers have proven adept at such classification tasks, the issue of neural network models often impedes their practical deployment in real-world settings. We present a novel graph-based multi-turn dialogue system called , which identifies a user's intent by identifying intent elements and a standard query from a dynamically constructed and extensible intent graph using reinforcement learning. In addition, we provide visualization components to monitor the immediate reasoning path for each turn of a dialogue, which greatly facilitates further improvement of the system.
Estimating the speed of vehicles using traffic cameras is a crucial task for traffic surveillance and management, enabling more optimal traffic flow, improved road safety, and lower environmental impact. Transportation-dependent systems, such as for navigation and logistics, have great potential to benefit from reliable speed estimation. While there is prior research in this area reporting competitive accuracy levels, their solutions lack reproducibility and robustness across different datasets. To address this, we provide a novel framework for automatic real-time vehicle speed calculation, which copes with more diverse data from publicly available traffic cameras to achieve greater robustness. Our model employs novel techniques to estimate the length of road segments via depth map prediction. Additionally, our framework is capable of handling realistic conditions such as camera movements and different video stream inputs automatically. We compare our model to three well-known models in the field using their benchmark datasets. While our model does not set a new state of the art regarding prediction performance, the results are competitive on realistic CCTV videos. At the same time, our end-to-end pipeline offers more consistent results, an easier implementation, and better compatibility. Its modular structure facilitates reproducibility and future improvements.
Large Language Models (LLMs) have taken Knowledge Representation -- and the world -- by storm. This inflection point marks a shift from explicit knowledge representation to a renewed focus on the hybrid representation of both explicit knowledge and parametric knowledge. In this position paper, we will discuss some of the common debate points within the community on LLMs (parametric knowledge) and Knowledge Graphs (explicit knowledge) and speculate on opportunities and visions that the renewed focus brings, as well as related research topics and challenges.
Given the success of Graph Neural Networks (GNNs) for structure-aware machine learning, numerous studies have explored their application to text classification, as an alternative to traditional feature representation models. However, most studies considered just a specific domain and validated on data with particular characteristics. This work presents an extensive empirical investigation of graph-based text representation methods proposed for text classification, identifying practical implications and open challenges in the field. We compare several GNN architectures as well as BERT across five datasets, encompassing short and also long documents. The results show that: i) graph performance is highly related to the textual input features and domain, ii) despite its outstanding performance, BERT has difficulties converging when dealing with short texts, iii) graph methods are particularly beneficial for longer documents.
Using model weights pretrained on a high-resource language as a warm start can reduce the need for data and compute to obtain high-quality language models in low-resource languages. To accommodate the new language, the pretrained vocabulary and embeddings need to be adapted. Previous work on embedding initialization for such adapted vocabularies has mostly focused on monolingual source models. In this paper, we investigate the multilingual source model setting and propose FOCUS - Fast Overlapping Token Combinations Using Sparsemax, a novel embedding initialization method that outperforms previous work when adapting XLM-R. FOCUS represents newly added tokens as combinations of tokens in the overlap of the pretrained and new vocabularies. The overlapping tokens are selected based on semantic similarity in an auxiliary token embedding space. Our implementation of FOCUS is publicly available on GitHub.