We evaluate the use of the open-source Llama-2 model for generating well-known, high-performance computing kernels (e.g., AXPY, GEMV, GEMM) on different parallel programming models and languages (e.g., C++: OpenMP, OpenMP Offload, OpenACC, CUDA, HIP; Fortran: OpenMP, OpenMP Offload, OpenACC; Python: numpy, Numba, pyCUDA, cuPy; and Julia: Threads, CUDA.jl, AMDGPU.jl). We built upon our previous work that is based on the OpenAI Codex, which is a descendant of GPT-3, to generate similar kernels with simple prompts via GitHub Copilot. Our goal is to compare the accuracy of Llama-2 and our original GPT-3 baseline by using a similar metric. Llama-2 has a simplified model that shows competitive or even superior accuracy. We also report on the differences between these foundational large language models as generative AI continues to redefine human-computer interactions. Overall, Copilot generates codes that are more reliable but less optimized, whereas codes generated by Llama-2 are less reliable but more optimized when correct.

We explore the novel application of Large Language Models to code optimization. We present a 7B-parameter transformer model trained from scratch to optimize LLVM assembly for code size. The model takes as input unoptimized assembly and outputs a list of compiler options to best optimize the program. Crucially, during training, we ask the model to predict the instruction counts before and after optimization, and the optimized code itself. These auxiliary learning tasks significantly improve the optimization performance of the model and improve the model's depth of understanding. We evaluate on a large suite of test programs. Our approach achieves a 3.0% improvement in reducing instruction counts over the compiler, outperforming two state-of-the-art baselines that require thousands of compilations. Furthermore, the model shows surprisingly strong code reasoning abilities, generating compilable code 91% of the time and perfectly emulating the output of the compiler 70% of the time.

This allows us to predict future actions accurately, video understanding, including video production and editing particularly from short-horizon observations - a key aspect [20, 41, 6, 38]. This work focuses on anticipating actions that prior works [11, 1, 22, 34, 2, 17] in action anticipation from short video segments and provides potential avenues fail to cater to. to enhance the editing process. In particular, the ability In our work, we utilize Knowledge Graphs (KG) to capture to extract actions from a video segment can be utilized the relationship between entities present in the video in two manners: 1) It allows for intelligent clip suggestions and link them to their respective affordances and the potential for future editing, namely the ability to suggest videos given tools that could be used to afford them in a particular what will likely happen next, and 2) it provides information way. Prior work [40, 26, 23, 15] has introduced efficient on what generally would happen, which allows editors to methods of identifying such relationships, which can refine their composition to either confirm or contradict a subsequently be utilized to identify the potential for certain viewer's expectation.

Hallucination in a foundation model (FM) refers to the generation of content that strays from factual reality or includes fabricated information. This survey paper provides an extensive overview of recent efforts that aim to identify, elucidate, and tackle the problem of hallucination, with a particular focus on ``Large'' Foundation Models (LFMs). The paper classifies various types of hallucination phenomena that are specific to LFMs and establishes evaluation criteria for assessing the extent of hallucination. It also examines existing strategies for mitigating hallucination in LFMs and discusses potential directions for future research in this area. Essentially, the paper offers a comprehensive examination of the challenges and solutions related to hallucination in LFMs.

We introduce SAGE; a Generative LLM for inferring attribute values for products across world-wide e-Commerce catalogs. We introduce a novel formulation of the attribute-value prediction problem as a Seq2Seq summarization task, across languages, product types and target attributes. Our novel modeling approach lifts the restriction of predicting attribute values within a pre-specified set of choices, as well as, the requirement that the sought attribute values need to be explicitly mentioned in the text. SAGE can infer attribute values even when such values are mentioned implicitly using periphrastic language, or not-at-all--as is the case for common-sense defaults. Additionally, SAGE is capable of predicting whether an attribute is inapplicable for the product at hand, or non-obtainable from the available information. SAGE is the first method able to tackle all aspects of the attribute-value-prediction task as they arise in practical settings in e-Commerce catalogs. A comprehensive set of experiments demonstrates the effectiveness of the proposed approach, as well as, its superiority against state-of-the-art competing alternatives. Moreover, our experiments highlight SAGE's ability to tackle the task of predicting attribute values in zero-shot setting; thereby, opening up opportunities for significantly reducing the overall number of labeled examples required for training.

This paper investigates the strategic decision-making capabilities of three Large Language Models (LLMs): GPT-3.5, GPT-4, and LLaMa-2, within the framework of game theory. Utilizing four canonical two-player games -- Prisoner's Dilemma, Stag Hunt, Snowdrift, and Prisoner's Delight -- we explore how these models navigate social dilemmas, situations where players can either cooperate for a collective benefit or defect for individual gain. Crucially, we extend our analysis to examine the role of contextual framing, such as diplomatic relations or casual friendships, in shaping the models' decisions. Our findings reveal a complex landscape: while GPT-3.5 is highly sensitive to contextual framing, it shows limited ability to engage in abstract strategic reasoning. Both GPT-4 and LLaMa-2 adjust their strategies based on game structure and context, but LLaMa-2 exhibits a more nuanced understanding of the games' underlying mechanics. These results highlight the current limitations and varied proficiencies of LLMs in strategic decision-making, cautioning against their unqualified use in tasks requiring complex strategic reasoning.

Transformers have become the dominant model in deep learning, but the reason for their superior performance is poorly understood. Here, we hypothesize that the strong performance of Transformers stems from an architectural bias towards mesa-optimization, a learned process running within the forward pass of a model consisting of the following two steps: (i) the construction of an internal learning objective, and (ii) its corresponding solution found through optimization. To test this hypothesis, we reverse-engineer a series of autoregressive Transformers trained on simple sequence modeling tasks, uncovering underlying gradient-based mesa-optimization algorithms driving the generation of predictions. Moreover, we show that the learned forward-pass optimization algorithm can be immediately repurposed to solve supervised few-shot tasks, suggesting that mesa-optimization might underlie the in-context learning capabilities of large language models. Finally, we propose a novel self-attention layer, the mesa-layer, that explicitly and efficiently solves optimization problems specified in context. We find that this layer can lead to improved performance in synthetic and preliminary language modeling experiments, adding weight to our hypothesis that mesa-optimization is an important operation hidden within the weights of trained Transformers.

Over the past two decades, dialogue modeling has made significant strides, moving from simple rule-based responses to personalized and persuasive response generation. However, despite these advancements, the objective functions and evaluation metrics for dialogue generation have remained stagnant, i.e., cross-entropy and BLEU, respectively. These lexical-based metrics have the following key limitations: (a) word-to-word matching without semantic consideration: It assigns the same credit for failure to generate 'nice' and 'rice' for 'good'. (b) missing context attribute for evaluating the generated response: Even if a generated response is relevant to the ongoing dialogue context, it may still be penalized for not matching the gold utterance provided in the corpus. In this paper, we first investigate these limitations comprehensively and propose a new loss function called Semantic Infused Contextualized diaLogue (SemTextualLogue) loss function. Furthermore, we formulate a new evaluation metric called Dialuation, which incorporates both context relevance and semantic appropriateness while evaluating a generated response. We conducted experiments on two benchmark dialogue corpora, encompassing both task-oriented and open-domain scenarios. We found that the dialogue generation model trained with SemTextualLogue loss attained superior performance (in both quantitative and qualitative evaluation) compared to the traditional cross-entropy loss function across the datasets and evaluation metrics.

In this work, we use large language models (LLMs) to augment and accelerate research on the P versus NP problem, one of the most important open problems in theoretical computer science and mathematics. Specifically, we propose Socratic reasoning, a general framework that promotes in-depth thinking with LLMs for complex problem-solving. Socratic reasoning encourages LLMs to recursively discover, solve, and integrate problems while facilitating self-evaluation and refinement. Our pilot study on the P vs. NP problem shows that GPT-4 successfully produces a proof schema and engages in rigorous reasoning throughout 97 dialogue turns, concluding "P $\neq$ NP", which is in alignment with (Xu and Zhou, 2023). The investigation uncovers novel insights within the extensive solution space of LLMs, shedding light on LLM for Science.

Inductive reasoning is a core problem-solving capacity: humans can identify underlying principles from a few examples, which can then be robustly generalized to novel scenarios. Recent work has evaluated large language models (LLMs) on inductive reasoning tasks by directly prompting them yielding "in context learning." This can work well for straightforward inductive tasks, but performs very poorly on more complex tasks such as the Abstraction and Reasoning Corpus (ARC). In this work, we propose to improve the inductive reasoning ability of LLMs by generating explicit hypotheses at multiple levels of abstraction: we prompt the LLM to propose multiple abstract hypotheses about the problem, in natural language, then implement the natural language hypotheses as concrete Python programs. These programs can be directly verified by running on the observed examples and generalized to novel inputs. Because of the prohibitive cost of generation with state-of-the-art LLMs, we consider a middle step to filter the set of hypotheses that will be implemented into programs: we either ask the LLM to summarize into a smaller set of hypotheses, or ask human annotators to select a subset of the hypotheses. We verify our pipeline's effectiveness on the ARC visual inductive reasoning benchmark, its variant 1D-ARC, and string transformation dataset SyGuS. On a random 40-problem subset of ARC, our automated pipeline using LLM summaries achieves 27.5% accuracy, significantly outperforming the direct prompting baseline (accuracy of 12.5%). With the minimal human input of selecting from LLM-generated candidates, the performance is boosted to 37.5%. (And we argue this is a lower bound on the performance of our approach without filtering.) Our ablation studies show that abstract hypothesis generation and concrete program representations are both beneficial for LLMs to perform inductive reasoning tasks.