This paper introduces MMMU-Pro, a robust version of the Massive Multi-discipline Multimodal Understanding and Reasoning (MMMU) benchmark. MMMU-Pro rigorously assesses multimodal models' true understanding and reasoning capabilities through a three-step process based on MMMU: (1) filtering out questions answerable by text-only models, (2) augmenting candidate options, and (3) introducing a vision-only input setting where questions are embedded within images. This setting challenges AI to truly "see" and "read" simultaneously, testing a fundamental human cognitive skill of seamlessly integrating visual and textual information. Results show that model performance is substantially lower on MMMU-Pro than on MMMU, ranging from 16.8% to 26.9% across models. We explore the impact of OCR prompts and Chain of Thought (CoT) reasoning, finding that OCR prompts have minimal effect while CoT generally improves performance. MMMU-Pro provides a more rigorous evaluation tool, closely mimicking real-world scenarios and offering valuable directions for future research in multimodal AI.

The rapid advancement in large language models (LLMs) comes with a significant increase in their parameter size, presenting challenges for adaptation and fine-tuning. Parameter-efficient fine-tuning (PEFT) methods are widely used to adapt LLMs for downstream tasks efficiently. In this paper, we propose Singular Values and Orthonormal Regularized Singular Vectors Adaptation, or SORSA, a novel PEFT method. We introduce a method to analyze the variation of the parameters by performing singular value decomposition (SVD) and discuss and analyze SORSA's superiority in minimizing the alteration in the SVD aspect. Each SORSA adapter consists of two main parts: trainable principal singular weights $W_p = U_p \Sigma_p V^\top_p$, and frozen residual weights $W_r = U_r \Sigma_r V^\top_r$. These parts are initialized by performing SVD on pre-trained weights. Moreover, we implement and analyze an orthonormal regularizer, which could effectively transfer the scaling information into $\Sigma_p$ and ultimately allows the training process to be more efficient. SORSA adapters could be merged during inference, thus eliminating any inference latency. After all, SORSA shows a faster convergence than PiSSA and LoRA in our experiments. On the MATH benchmark, Llama 2 7B adapted using SORSA achieved 10.36% accuracy, outperforming LoRA (5.50%), Full FT (7.22%), and PiSSA (7.44%). On the GSM-8K benchmark, SORSA achieved 56.03% accuracy, surpassing LoRA (42.30%), Full FT (49.05%), and PiSSA (53.07%). We conclude that SORSA offers a new perspective on parameter-efficient fine-tuning, demonstrating remarkable performance. The code is available at https://github.com/Gunale0926/SORSA.

Human-friendly natural language has been explored for tasks such as driving explanation and 3D captioning. However, previous works primarily focused on the paradigm of declarative interpretability, where the natural language interpretations are not grounded in the intermediate outputs of AD systems, making the interpretations only declarative. In contrast, aligned interpretability establishes a connection between language and the intermediate outputs of AD systems. Here we introduce Hint-AD, an integrated AD-language system that generates language aligned with the holistic perceptionprediction-planning outputs of the AD model. By incorporating the intermediate outputs and a holistic token mixer sub-network for effective feature adaptation, Hint-AD achieves desirable accuracy, achieving state-of-the-art results in driving language tasks including driving explanation, 3D dense captioning, and command prediction. To facilitate further study on driving explanation task on nuScenes, we also introduce a human-labeled dataset, Nu-X.

The increasing amount of published scholarly articles, exceeding 2.5 million yearly, raises the challenge for researchers in following scientific progress. Integrating the contributions from scholarly articles into a novel type of cognitive knowledge graph (CKG) will be a crucial element for accessing and organizing scholarly knowledge, surpassing the insights provided by titles and abstracts. This research focuses on effectively conveying structured scholarly knowledge by utilizing large language models (LLMs) to categorize scholarly articles and describe their contributions in a structured and comparable manner. While previous studies explored language models within specific research domains, the extensive domain-independent knowledge captured by LLMs offers a substantial opportunity for generating structured contribution descriptions as CKGs. Additionally, LLMs offer customizable pathways through prompt engineering or fine-tuning, thus facilitating to leveraging of smaller LLMs known for their efficiency, cost-effectiveness, and environmental considerations. Our methodology involves harnessing LLM knowledge, and complementing it with domain expert-verified scholarly data sourced from a CKG. This strategic fusion significantly enhances LLM performance, especially in tasks like scholarly article categorization and predicate recommendation. Our method involves fine-tuning LLMs with CKG knowledge and additionally injecting knowledge from a CKG with a novel prompting technique significantly increasing the accuracy of scholarly knowledge extraction. We integrated our approach in the Open Research Knowledge Graph (ORKG), thus enabling precise access to organized scholarly knowledge, crucially benefiting domain-independent scholarly knowledge exchange and dissemination among policymakers, industrial practitioners, and the general public.

Jailbreak vulnerabilities in Large Language Models (LLMs) refer to methods that extract malicious content from the model by carefully crafting prompts or suffixes, which has garnered significant attention from the research community. However, traditional attack methods, which primarily focus on the semantic level, are easily detected by the model. These methods overlook the difference in the model's alignment protection capabilities at different output stages. To address this issue, we propose an adaptive position pre-fill jailbreak attack approach for executing jailbreak attacks on LLMs. Our method leverages the model's instruction-following capabilities to first output pre-filled safe content, then exploits its narrative-shifting abilities to generate harmful content. Extensive black-box experiments demonstrate our method can improve the attack success rate by 47% on the widely recognized secure model (Llama2) compared to existing approaches. Our code can be found at: https://github.com/Yummy416/AdaPPA.

Traditional dialogue state tracking approaches heavily rely on extensive training data and handcrafted features, limiting their scalability and adaptability to new domains. In this paper, we propose a novel method that leverages inference and in-context learning with ChatGPT for domain transfer in dialogue state tracking, without any parameter updates. By guiding ChatGPT's chain of thought, we enable it to retrieve relevant examples and generalize knowledge to accurately infer dialogue states, solely through inference. Experimental results on the MultiWOZ dataset demonstrate competitive performance and promising generalization across domains. Our parameter-free approach offers a scalable and adaptable solution, opening new research directions in domain transfer learning.

Retrieval-Augmented Generation (RAG) has emerged as a common paradigm to use Large Language Models (LLMs) alongside private and up-to-date knowledge bases. In this work, we address the challenges of using LLM-as-a-Judge when evaluating grounded answers generated by RAG systems. To assess the calibration and discrimination capabilities of judge models, we identify 7 generator failure modes and introduce GroUSE (Grounded QA Unitary Scoring of Evaluators), a meta-evaluation benchmark of 144 unit tests. This benchmark reveals that existing automated RAG evaluation frameworks often overlook important failure modes, even when using GPT-4 as a judge. To improve on the current design of automated RAG evaluation frameworks, we propose a novel pipeline and find that while closed models perform well on GroUSE, state-of-the-art open-source judges do not generalize to our proposed criteria, despite strong correlation with GPT-4's judgement. Our findings suggest that correlation with GPT-4 is an incomplete proxy for the practical performance of judge models and should be supplemented with evaluations on unit tests for precise failure mode detection. We further show that finetuning Llama-3 on GPT-4's reasoning traces significantly boosts its evaluation capabilities, improving upon both correlation with GPT-4's evaluations and calibration on reference situations.

We present a toolkit for creating low-cost Mixture-of-Domain-Experts (MOE) from trained models. The toolkit can be used for creating a mixture from models or from adapters. We perform extensive tests and offer guidance on defining the architecture of the resulting MOE using the toolkit. A public repository is available.

Evaluating the quality of text generated by large language models (LLMs) remains a significant challenge. Traditional metrics often fail to align well with human judgments, particularly in tasks requiring creativity and nuance. In this paper, we propose \textsc{Check-Eval}, a novel evaluation framework leveraging LLMs to assess the quality of generated text through a checklist-based approach. \textsc{Check-Eval} can be employed as both a reference-free and reference-dependent evaluation method, providing a structured and interpretable assessment of text quality. The framework consists of two main stages: checklist generation and checklist evaluation. We validate \textsc{Check-Eval} on two benchmark datasets: Portuguese Legal Semantic Textual Similarity and \textsc{SummEval}. Our results demonstrate that \textsc{Check-Eval} achieves higher correlations with human judgments compared to existing metrics, such as \textsc{G-Eval} and \textsc{GPTScore}, underscoring its potential as a more reliable and effective evaluation framework for natural language generation tasks. The code for our experiments is available at \url{https://anonymous.4open.science/r/check-eval-0DB4}

In the realm of Large Language Models (LLMs), the ability to process long contexts is increasingly crucial for tasks such as multi-round dialogues, code generation, and document summarization. This paper addresses the challenges of enhancing the long-context performance, reducing computational complexity, and leveraging pretrained models collectively termed the "impossible triangle." We introduce E2LLM (Encoder Elongated Large Language Models), a novel approach that effectively navigates this paradox. The method involves splitting long contexts into chunks, compressing each into embedding vectors via a pretrained text encoder, and utilizing an adapter to align these representations with a decoder-only LLM. Two training objectives, focusing on reconstruction of the encoder output and long-context instruction fine-tuning, are employed to facilitate the understanding of soft prompts by the LLM. Experimental results demonstrate that E2LLM achieves superior performance in long-context scenarios while balancing efficiency, performance, and compatibility with pretrained models. Our framework thus represents a significant advancement in the field, contributing to effective long-text modeling.