Large Language Models (LLMs) for code are a family of high-parameter, transformer-based neural networks pre-trained on massive datasets of both natural and programming languages. These models are rapidly being employed in commercial AI-based developer tools, such as GitHub CoPilot. However, measuring and explaining their effectiveness on programming tasks is a challenging proposition, given their size and complexity. The methods for evaluating and explaining LLMs for code are inextricably linked. That is, in order to explain a model's predictions, they must be reliably mapped to fine-grained, understandable concepts. Once this mapping is achieved, new methods for detailed model evaluations are possible. However, most current explainability techniques and evaluation benchmarks focus on model robustness or individual task performance, as opposed to interpreting model predictions. To this end, this paper introduces ASTxplainer, an explainability method specific to LLMs for code that enables both new methods for LLM evaluation and visualizations of LLM predictions that aid end-users in understanding model predictions. At its core, ASTxplainer provides an automated method for aligning token predictions with AST nodes, by extracting and aggregating normalized model logits within AST structures. To demonstrate the practical benefit of ASTxplainer, we illustrate the insights that our framework can provide by performing an empirical evaluation on 12 popular LLMs for code using a curated dataset of the most popular GitHub projects. Additionally, we perform a user study examining the usefulness of an ASTxplainer-derived visualization of model predictions aimed at enabling model users to explain predictions. The results of these studies illustrate the potential for ASTxplainer to provide insights into LLM effectiveness, and aid end-users in understanding predictions.

Large language models (LLMs) are incredibly powerful at comprehending and generating data in the form of text, but are brittle and error-prone. There has been an advent of toolkits and recipes centered around so-called prompt engineering-the process of asking an LLM to do something via a series of prompts. However, for LLM-powered data processing workflows, in particular, optimizing for quality, while keeping cost bounded, is a tedious, manual process. We put forth a vision for declarative prompt engineering. We view LLMs like crowd workers and leverage ideas from the declarative crowdsourcing literature-including leveraging multiple prompting strategies, ensuring internal consistency, and exploring hybrid-LLM-non-LLM approaches-to make prompt engineering a more principled process. Preliminary case studies on sorting, entity resolution, and imputation demonstrate the promise of our approach

The misuse of large language models (LLMs) has garnered significant attention from the general public and LLM vendors. In response, efforts have been made to align LLMs with human values and intent use. However, a particular type of adversarial prompts, known as jailbreak prompt, has emerged and continuously evolved to bypass the safeguards and elicit harmful content from LLMs. In this paper, we conduct the first measurement study on jailbreak prompts in the wild, with 6,387 prompts collected from four platforms over six months. Leveraging natural language processing technologies and graph-based community detection methods, we discover unique characteristics of jailbreak prompts and their major attack strategies, such as prompt injection and privilege escalation. We also observe that jailbreak prompts increasingly shift from public platforms to private ones, posing new challenges for LLM vendors in proactive detection. To assess the potential harm caused by jailbreak prompts, we create a question set comprising 46,800 samples across 13 forbidden scenarios. Our experiments show that current LLMs and safeguards cannot adequately defend jailbreak prompts in all scenarios. Particularly, we identify two highly effective jailbreak prompts which achieve 0.99 attack success rates on ChatGPT (GPT-3.5) and GPT-4, and they have persisted online for over 100 days. Our work sheds light on the severe and evolving threat landscape of jailbreak prompts. We hope our study can facilitate the research community and LLM vendors in promoting safer and regulated LLMs.

Despite speculation that recent large language models (LLMs) are likely to be used maliciously to improve the quality or scale of influence operations, uncertainty persists regarding the economic value that LLMs offer propagandists. This research constructs a model of costs facing propagandists for content generation at scale and analyzes (1) the potential savings that LLMs could offer propagandists, (2) the potential deterrent effect of monitoring controls on API-accessible LLMs, and (3) the optimal strategy for propagandists choosing between multiple private and/or open source LLMs when conducting influence operations. Primary results suggest that LLMs need only produce usable outputs with relatively low reliability (roughly 25%) to offer cost savings to propagandists, that the potential reduction in content generation costs can be quite high (up to 70% for a highly reliable model), and that monitoring capabilities have sharply limited cost imposition effects when alternative open source models are available. In addition, these results suggest that nation-states -- even those conducting many large-scale influence operations per year -- are unlikely to benefit economically from training custom LLMs specifically for use in influence operations.

Recent advancements in Large Vision-Language Models (LVLMs) have demonstrated significant progress in tackling complex multimodal tasks. Among these cutting-edge developments, Google's Bard stands out for its remarkable multimodal capabilities, promoting comprehensive comprehension and reasoning across various domains. This work presents an early and holistic evaluation of LVLMs' multimodal abilities, with a particular focus on Bard, by proposing a lightweight variant of LVLM-eHub, named Tiny LVLM-eHub. In comparison to the vanilla version, Tiny LVLM-eHub possesses several appealing properties. Firstly, it provides a systematic assessment of six categories of multimodal capabilities, including visual perception, visual knowledge acquisition, visual reasoning, visual commonsense, object hallucination, and embodied intelligence, through quantitative evaluation of $42$ standard text-related visual benchmarks. Secondly, it conducts an in-depth analysis of LVLMs' predictions using the ChatGPT Ensemble Evaluation (CEE), which leads to a robust and accurate evaluation and exhibits improved alignment with human evaluation compared to the word matching approach. Thirdly, it comprises a mere $2.1$K image-text pairs, facilitating ease of use for practitioners to evaluate their own offline LVLMs. Through extensive experimental analysis, this study demonstrates that Bard outperforms previous LVLMs in most multimodal capabilities except object hallucination, to which Bard is still susceptible. Tiny LVLM-eHub serves as a baseline evaluation for various LVLMs and encourages innovative strategies aimed at advancing multimodal techniques. Our project is publicly available at \url{https://github.com/OpenGVLab/Multi-Modality-Arena}.

Transformer architectures and models have made significant progress in language-based tasks. In this area, is BERT one of the most widely used and freely available transformer architecture. In our work, we use BERT for context-based phrase recognition of magic spells in the Harry Potter novel series. Spells are a common part of active magic in fantasy novels. Typically, spells are used in a specific context to achieve a supernatural effect. A series of investigations were conducted to see if a Transformer architecture could recognize such phrases based on their context in the Harry Potter saga. For our studies a pre-trained BERT model was used and fine-tuned utilising different datasets and training methods to identify the searched context. By considering different approaches for sequence classification as well as token classification, it is shown that the context of spells can be recognised. According to our investigations, the examined sequence length for fine-tuning and validation of the model plays a significant role in context recognition. Based on this, we have investigated whether spells have overarching properties that allow a transfer of the neural network models to other fantasy universes as well. The application of our model showed promising results and is worth to be deepened in subsequent studies.

Large language models (LLMs) have demonstrated remarkable performance in a wide range of natural language tasks. However, as these models continue to grow in size, they face significant challenges in terms of computational costs. Additionally, LLMs often lack efficient domain-specific understanding, which is particularly crucial in specialized fields such as aviation and healthcare. To boost the domain-specific understanding, we propose, KITLM, a novel knowledge base integration approach into language model through relevant information infusion. By integrating pertinent knowledge, not only the performance of the language model is greatly enhanced, but the model size requirement is also significantly reduced while achieving comparable performance. Our proposed knowledge-infused model surpasses the performance of both GPT-3.5-turbo and the state-of-the-art knowledge infusion method, SKILL, achieving over 1.5 times improvement in exact match scores on the MetaQA. KITLM showed a similar performance boost in the aviation domain with AeroQA. The drastic performance improvement of KITLM over the existing methods can be attributed to the infusion of relevant knowledge while mitigating noise. In addition, we release two curated datasets to accelerate knowledge infusion research in specialized fields: a) AeroQA, a new benchmark dataset designed for multi-hop question-answering within the aviation domain, and b) Aviation Corpus, a dataset constructed from unstructured text extracted from the National Transportation Safety Board reports. Our research contributes to advancing the field of domain-specific language understanding and showcases the potential of knowledge infusion techniques in improving the performance of language models on question-answering.

The Machine Learning as a Service (MLaaS) market is rapidly expanding and becoming more mature. For example, OpenAI's ChatGPT is an advanced large language model (LLM) that generates responses for various queries with associated fees. Although these models can deliver satisfactory performance, they are far from perfect. Researchers have long studied the vulnerabilities and limitations of LLMs, such as adversarial attacks and model toxicity. Inevitably, commercial ML models are also not exempt from such issues, which can be problematic as MLaaS continues to grow. In this paper, we discover a new attack strategy against LLM APIs, namely the prompt abstraction attack. Specifically, we propose Mondrian, a simple and straightforward method that abstracts sentences, which can lower the cost of using LLM APIs. In this approach, the adversary first creates a pseudo API (with a lower established price) to serve as the proxy of the target API (with a higher established price). Next, the pseudo API leverages Mondrian to modify the user query, obtain the abstracted response from the target API, and forward it back to the end user. Our results show that Mondrian successfully reduces user queries' token length ranging from 13% to 23% across various tasks, including text classification, generation, and question answering. Meanwhile, these abstracted queries do not significantly affect the utility of task-specific and general language models like ChatGPT. Mondrian also reduces instruction prompts' token length by at least 11% without compromising output quality. As a result, the prompt abstraction attack enables the adversary to profit without bearing the cost of API development and deployment.

In this work, we present SciGraphQA, a synthetic multi-turn question-answer dataset related to academic graphs. SciGraphQA is 13 times larger than ChartVQA, the previously largest chart-visual question-answering dataset. It is also the largest open-sourced chart VQA dataset with non-synthetic charts. To build our dataset, we selected 290,000 Computer Science or Machine Learning ArXiv papers published between 2010 and 2020, and then used Palm-2 to generate 295K samples of open-vocabulary multi-turn question-answering dialogues about the graphs. As context, we provided the text-only Palm-2 with paper title, abstract, paragraph mentioning the graph, and rich text contextual data from the graph itself, obtaining dialogues with an average 2.23 question-answer turns for each graph. We asked GPT-4 to assess the matching quality of our question-answer turns given the paper's context, obtaining an average rating of 8.7/10 on our 3K test set. We evaluated the 0-shot capability of the most popular MLLM models such as LLaVa, mPLUGowl, BLIP-2, and openFlamingo's on our dataset, finding LLaVA-13B being the most performant with a CIDEr score of 0.08. We further enriched the question prompts for LLAVA by including the serialized data tables extracted from the graphs using the DePlot model, boosting LLaVA's 0-shot CIDEr to 0.15. To verify the validity of our dataset, we also fine-tuned LLaVa using our dataset, reaching a substantially higher CIDEr score of 0.26. We anticipate further accuracy improvement by including segmentation mask tokens and leveraging larger LLM backbones coupled with emergent prompting techniques. Our code and data are open-sourced.

The low-rank adaptation (LoRA) method can largely reduce the amount of trainable parameters for fine-tuning large language models (LLMs), however, it still requires expensive activation memory to update low-rank weights. Reducing the number of LoRA layers or using activation recomputation could harm the finetuning performance or increase the computational overhead. In this work, we present LoRA-FA, a memory-efficient fine-tuning method that reduces the activation memory without performance degradation and expensive recomputation. LoRA-FA chooses to freeze the projection-down weight of A and update the projection-up weight of B in each LoRA layer. It ensures the change of model weight reside in a low-rank space during LLMs fine-tuning, while eliminating the requirement to store full-rank input activations. We conduct extensive experiments across multiple model types (RoBERTa, T5, LLaMA) and model scales. Our results show that LoRA-FA can always achieve close fine-tuning accuracy across different tasks compared to full parameter fine-tuning and LoRA. Furthermore, LoRA-FA can reduce the overall memory cost by up to 1.4 compared to LoRA. However, fine-tuning LLMs with full parameter is prohibitively expensive, for example, fine-tuning a LLaMA-65B (Touvron et al., 2023a) model with AdamW (Loshchilov & Hutter, 2017) requires more than 1TB of GPU memory to store model parameter, gradient, and optimizer states (Rajbhandari et al., 2020). To reduce the memory of full-parameter fine-tuning, parameter-efficient fine-tuning (PEFT) methods are proposed to update only a small fraction of parameters, such as adapter weights (Houlsby et al., 2019; Hu et al., 2022) and prompt weights (Li & Liang, 2021; Lester et al., 2021).