The recent surge of open-source large language models (LLMs) enables developers to create AI-based solutions while maintaining control over aspects such as privacy and compliance, thereby providing governance and ownership of the model deployment process. To utilize these LLMs, inference engines are needed. These engines load the model's weights onto available resources, such as GPUs, and process queries to generate responses. The speed of inference, or performance, of the LLM, is critical for real-time applications, as it computes millions or billions of floating point operations per inference. Recently, advanced inference engines such as vLLM have emerged, incorporating novel mechanisms such as efficient memory management to achieve state-of-the-art performance. In this paper, we analyze the performance, particularly the throughput (tokens generated per unit of time), of 20 LLMs using two inference libraries: vLLM and HuggingFace's pipelines. We investigate how various hyperparameters, which developers must configure, influence inference performance. Our results reveal that throughput landscapes are irregular, with distinct peaks, highlighting the importance of hyperparameter optimization to achieve maximum performance. We also show that applying hyperparameter optimization when upgrading or downgrading the GPU model used for inference can improve throughput from HuggingFace pipelines by an average of 9.16% and 13.7%, respectively.

General-purpose object placement is a fundamental capability of an intelligent generalist robot, i.e., being capable of rearranging objects following human instructions even in novel environments. To achieve this, we break the rearrangement down into three parts, including object localization, goal imagination and robot control, and propose a framework named SPORT. SPORT leverages pre-trained large vision models for broad semantic reasoning about objects, and learns a diffusion-based 3D pose estimator to ensure physically-realistic results. Only object types (to be moved or reference) are communicated between these two parts, which brings two benefits. One is that we can fully leverage the powerful ability of open-set object localization and recognition since no specific fine-tuning is needed for robotic scenarios. Furthermore, the diffusion-based estimator only need to "imagine" the poses of the moving and reference objects after the placement, while no necessity for their semantic information. Thus the training burden is greatly reduced and no massive training is required. The training data for goal pose estimation is collected in simulation and annotated with GPT-4. A set of simulation and real-world experiments demonstrate the potential of our approach to accomplish general-purpose object rearrangement, placing various objects following precise instructions.

The rise of large language models (LLMs) has enabled us to seek answers to inherently debatable questions on LLM chatbots, necessitating a reliable way to evaluate their ability. However, traditional QA benchmarks assume fixed answers are inadequate for this purpose. To address this, we introduce DebateQA, a dataset of 2,941 debatable questions, each accompanied by multiple human-annotated partial answers that capture a variety of perspectives. We develop two metrics: Perspective Diversity, which evaluates the comprehensiveness of perspectives, and Dispute Awareness, which assesses if the LLM acknowledges the question's debatable nature. Experiments demonstrate that both metrics align with human preferences and are stable across different underlying models. Using DebateQA with two metrics, we assess 12 popular LLMs and retrieval-augmented generation methods. Our findings reveal that while LLMs generally excel at recognizing debatable issues, their ability to provide comprehensive answers encompassing diverse perspectives varies considerably.

Learning token embeddings based on token co-occurrence statistics has proven effective for both pre-training and fine-tuning in natural language processing. However, recent studies have pointed out the distribution of learned embeddings degenerates into anisotropy, and even pre-trained language models (PLMs) suffer from a loss of semantics-related information in embeddings for low-frequency tokens. This study first analyzes fine-tuning dynamics of a PLM, BART-large, and demonstrates its robustness against degeneration. On the basis of this finding, we propose DefinitionEMB, a method that utilizes definitions to construct isotropically distributed and semantics-related token embeddings for PLMs while maintaining original robustness during fine-tuning. Our experiments demonstrate the effectiveness of leveraging definitions from Wiktionary to construct such embeddings for RoBERTa-base and BART-large. Furthermore, the constructed embeddings for low-frequency tokens improve the performance of these models across various GLUE and four text summarization datasets.

Large language models (LLMs) are the result of a massive experiment in bottom-up, data-driven reverse engineering of language at scale. Despite their utility in a number of downstream NLP tasks, ample research has shown that LLMs are incapable of performing reasoning in tasks that require quantification over and the manipulation of symbolic variables (e.g., planning and general problem solving) - see for example [25][26]. In this document, however, we will focus on testing LLMs for their language understanding capabilities, their supposed forte. In this regard we believe that we have not been testing the language understanding capabilities of large language models (LLMs) properly. Prompting LLMs and asking for responses will always look impressive because that's how LLMs were designed, i.e., to generate text. The proper method of testing the understanding capabilities of LLMs, we argue, is to prompt LLMs in reverse: give the LLM a snippet of text and query their understanding of the input text by asking the LLM questions against the input text. As we will show here the language understanding capabilities of LLMs have been widely exaggerated. By testing the understanding capabilities properly - i.e., by giving the LLM snippets of text as input and then querying what the LLM "understood" it will become apparent that LLMs do not truly understand language, beyond very superficial inferences that are essentially the byproduct of the memorization of massive amounts of ingested text.

The LLM Agent, equipped with a code interpreter, is capable of automatically solving real-world coding tasks, such as data analysis and image editing. However, existing benchmarks primarily focus on either simplistic tasks, such as completing a few lines of code, or on extremely complex and specific tasks at the repository level, neither of which are representative of various daily coding tasks. To address this gap, we introduce \textbf{PyBench}, a benchmark encompassing five main categories of real-world tasks, covering more than 10 types of files. Given a high-level user query and related files, the LLM Agent needs to reason and execute Python code via a code interpreter for a few turns before making a formal response to fulfill the user's requirements. Successfully addressing tasks in PyBench demands a robust understanding of various Python packages, superior reasoning capabilities, and the ability to incorporate feedback from executed code. Our evaluations indicate that current open-source LLMs are struggling with these tasks. Hence, we conduct analysis and experiments on four kinds of datasets proving that comprehensive abilities are needed for PyBench. Our fine-tuned 8B size model: \textbf{PyLlama3} achieves an exciting performance on PyBench which surpasses many 33B and 70B size models. Our Benchmark, Training Dataset, and Model are available at: {https://github.com/Mercury7353/PyBench}

In online learning, the ability to provide quick and accurate feedback to learners is crucial. In skill-based learning, learners need to understand the underlying concepts and mechanisms of a skill to be able to apply it effectively. While videos are a common tool in online learning, they cannot comprehend or assess the skills being taught. Additionally, while Generative AI methods are effective in searching and retrieving answers from a text corpus, it remains unclear whether these methods exhibit any true understanding. This limits their ability to provide explanations of skills or help with problem-solving. This paper proposes a novel approach that merges Cognitive AI and Generative AI to address these challenges. We employ a structured knowledge representation, the TMK (Task-Method-Knowledge) model, to encode skills taught in an online Knowledge-based AI course. Leveraging techniques such as Large Language Models, Chain-of-Thought, and Iterative Refinement, we outline a framework for generating reasoned explanations in response to learners' questions about skills.

The Retrieval-Augmented Language Model (RALM) has shown remarkable performance on knowledge-intensive tasks by incorporating external knowledge during inference, which mitigates the factual hallucinations inherited in large language models (LLMs). Despite these advancements, challenges persist in the implementation of RALMs, particularly concerning their reliability and traceability. To be specific, the irrelevant document retrieval may result in unhelpful response generation or even deteriorate the performance of LLMs, while the lack of proper citations in generated outputs complicates efforts to verify the trustworthiness of the models. To this end, we propose a novel self-reasoning framework aimed at improving the reliability and traceability of RALMs, whose core idea is to leverage reasoning trajectories generated by the LLM itself. The framework involves constructing self-reason trajectories with three processes: a relevance-aware process, an evidence-aware selective process, and a trajectory analysis process. We have evaluated our framework across four public datasets (two short-form QA datasets, one long-form QA dataset, and one fact verification dataset) to demonstrate the superiority of our method, which can outperform existing state-of-art models and can achieve comparable performance with GPT-4, while only using 2,000 training samples.

Prompt tuning is a modular and efficient solution for training large language models (LLMs). One of its main advantages is task modularity, making it suitable for multi-task problems. However, current soft-prompt-based methods often sacrifice multi-task modularity, requiring the training process to be fully or partially repeated for each newly added task. While recent work on task vectors applied arithmetic operations on full model weights to achieve the desired multi-task performance, a similar approach for soft-prompts is still missing. To this end, we introduce Task Prompt Vectors, created by element-wise difference between weights of tuned soft-prompts and their random initialization. Experimental results on 12 NLU datasets show that task prompt vectors can be used in low-resource settings to effectively initialize prompt tuning on similar tasks. In addition, we show that task prompt vectors are independent of the random initialization of prompt tuning. This allows prompt arithmetics with the pre-trained vectors from different tasks. In this way, by arithmetic addition of task prompt vectors from multiple tasks, we are able to outperform a state-of-the-art baseline in some cases.

OpenAI will give the US AI Safety Institute early access to its next model as part of its safety efforts, Sam Altman has revealed in a tweet. Apparently, the company has been working with the consortium "to push forward the science of AI evaluations." The National Institute of Standards and Technology (NIST) has formally established the Artificial Intelligence Safety Institute earlier this year, though Vice President Kamala Harris announced it back in 2023 at the UK AI Safety Summit. Based on the NIST's description of the consortium, it's meant "to develop science-based and empirically backed guidelines and standards for AI measurement and policy, laying the foundation for AI safety across the world." The company, along with DeepMind, similarly pledged to share AI models with the UK government last year.