Current datasets for long-form video understanding often fall short of providing genuine long-form comprehension challenges, as many tasks derived from these datasets can be successfully tackled by analyzing just one or a few random frames from a video. To address this issue, we present a novel dataset and benchmark, CinePile, specifically designed for authentic long-form video understanding. This paper details our innovative approach for creating a question-answer dataset, utilizing advanced LLMs with human-in-the-loop and building upon human-generated raw data. Our comprehensive dataset comprises 305,000 multiple-choice questions (MCQs), covering various visual and multimodal aspects, including temporal comprehension, understanding human-object interactions, and reasoning about events or actions within a scene. Additionally, we evaluate recent video-centric LLMs, both open-source and proprietary, on the test split of our dataset. The findings reveal that even state-of-the-art video-centric LLMs significantly lag behind human performance in these tasks, highlighting the complexity and challenge inherent in video understanding.

We propose SLoPe, a Double-Pruned Sparse Plus Lazy Low-rank Adapter Pretraining method for LLMs that improves the accuracy of sparse LLMs while accelerating their pretraining and inference and reducing their memory footprint. Sparse pretraining of LLMs reduces the accuracy of the model, to overcome this, prior work uses dense models during fine-tuning. SLoPe improves the accuracy of sparsely pretrained models by adding low-rank adapters in the final 1% iterations of pretraining without adding significant overheads to the model pretraining and inference. In addition, SLoPe uses a double-pruned backward pass formulation that prunes the transposed weight matrix using N:M sparsity structures to enable an accelerated sparse backward pass. SLoPe accelerates the training and inference of models with billions of parameters up to $1.14\times$ and $1.34\times$ respectively (OPT-33B and OPT-66B) while reducing their memory usage by up to $0.77\times$ and $0.51\times$ for training and inference respectively.

Fine-tuning large language models (LLMs) requires significant memory, often exceeding the capacity of a single GPU. A common solution to this memory challenge is offloading compute and data from the GPU to the CPU. However, this approach is hampered by the limited bandwidth of commodity hardware, which constrains communication between the CPU and GPU. In this paper, we present an offloading framework, LSP_Offload, that enables near-native speed LLM fine-tuning on commodity hardware through learned subspace projectors. Our data-driven approach involves learning an efficient sparse compressor that minimizes communication with minimal precision loss. Additionally, we introduce a novel layer-wise communication schedule to maximize parallelism between communication and computation. As a result, our framework can fine-tune a 1.3 billion parameter model on a 4GB laptop GPU and a 7 billion parameter model on an NVIDIA RTX 4090 GPU with 24GB memory, achieving only a 31% slowdown compared to fine-tuning with unlimited memory. Compared to state-of-the-art offloading frameworks, our approach increases fine-tuning throughput by up to 3.33 times and reduces end-to-end fine-tuning time by 33.1%~62.5% when converging to the same accuracy.

The World Wide Web provides unrivalled access to information globally, including factual news reporting and commentary. However, state actors and commercial players increasingly spread biased (distorted) or fake (non-factual) information to promote their agendas. We compare several large, pre-trained language models on the task of sentence-level news bias detection and sub-type classification, providing quantitative and qualitative results. Our findings are to be seen as part of a wider effort towards realizing the conceptual vision, articulated by Fuhr et al. [10], of a "nutrition label" for online content for the social good.

Capturing molecular knowledge with representation learning approaches holds significant potential in vast scientific fields such as chemistry and life science. An effective and generalizable molecular representation is expected to capture the consensus and complementary molecular expertise from diverse views and perspectives. However, existing works fall short in learning multi-view molecular representations, due to challenges in explicitly incorporating view information and handling molecular knowledge from heterogeneous sources. To address these issues, we present MV-Mol, a molecular representation learning model that harvests multi-view molecular expertise from chemical structures, unstructured knowledge from biomedical texts, and structured knowledge from knowledge graphs. We utilize text prompts to model view information and design a fusion architecture to extract view-based molecular representations. We develop a two-stage pre-training procedure, exploiting heterogeneous data of varying quality and quantity. Through extensive experiments, we show that MV-Mol provides improved representations that substantially benefit molecular property prediction. Additionally, MV-Mol exhibits state-of-the-art performance in multi-modal comprehension of molecular structures and texts. Code and data are available at https://github.com/PharMolix/OpenBioMed.

The integration of Artificial Intelligence (AI), especially Large Language Models (LLMs), into the clinical diagnosis process offers significant potential to improve the efficiency and accessibility of medical care. While LLMs have shown some promise in the medical domain, their application in clinical diagnosis remains underexplored, especially in real-world clinical practice, where highly sophisticated, patient-specific decisions need to be made. Current evaluations of LLMs in this field are often narrow in scope, focusing on specific diseases or specialties and employing simplified diagnostic tasks. To bridge this gap, we introduce CliBench, a novel benchmark developed from the MIMIC IV dataset, offering a comprehensive and realistic assessment of LLMs' capabilities in clinical diagnosis. This benchmark not only covers diagnoses from a diverse range of medical cases across various specialties but also incorporates tasks of clinical significance: treatment procedure identification, lab test ordering and medication prescriptions. Supported by structured output ontologies, CliBench enables a precise and multi-granular evaluation, offering an in-depth understanding of LLM's capability on diverse clinical tasks of desired granularity. We conduct a zero-shot evaluation of leading LLMs to assess their proficiency in clinical decision-making. Our preliminary results shed light on the potential and limitations of current LLMs in clinical settings, providing valuable insights for future advancements in LLM-powered healthcare.

Verification-aware programming languages such as Dafny and F* provide means to formally specify and prove properties of programs. Although the problem of checking an implementation against a specification can be defined mechanically, there is no algorithmic way of ensuring the correctness of the user-intent formalization for programs -- that a specification adheres to the user's intent behind the program. The intent or requirement is expressed informally in natural language and the specification is a formal artefact. The advent of large language models (LLMs) has made strides bridging the gap between informal intent and formal program implementations recently, driven in large parts due to benchmarks and automated metrics for evaluation. Recent work has proposed evaluating {\it user-intent formalization} problem for mainstream programming languages~\cite{endres-fse24}. However, such an approach does not readily extend to verification-aware languages that support rich specifications (containing quantifiers and ghost variables) that cannot be evaluated through dynamic execution. Previous work also required generating program mutants using LLMs to create the benchmark. We advocate an alternate approach of {\it symbolically testing specifications} to provide an intuitive metric for evaluating the quality of specifications for verification-aware languages. We demonstrate that our automated metric agrees closely with mostly GPT-4 generated and human-labeled dataset of roughly 150 Dafny specifications for the popular MBPP code-generation benchmark, yet demonstrates cases where the human labeling is not perfect. We believe our work provides a stepping stone to enable the establishment of a benchmark and research agenda for the problem of user-intent formalization for programs.

In this report, we introduce the Gemini 1.5 family of models, representing the next generation of highly compute-efficient multimodal models capable of recalling and reasoning over fine-grained information from millions of tokens of context, including multiple long documents and hours of video and audio. The family includes two new models: (1) an updated Gemini 1.5 Pro, which exceeds the February version on the great majority of capabilities and benchmarks; (2) Gemini 1.5 Flash, a more lightweight variant designed for efficiency with minimal regression in quality. Gemini 1.5 models achieve near-perfect recall on long-context retrieval tasks across modalities, improve the state-of-the-art in long-document QA, long-video QA and long-context ASR, and match or surpass Gemini 1.0 Ultra's state-of-the-art performance across a broad set of benchmarks. Studying the limits of Gemini 1.5's long-context ability, we find continued improvement in next-token prediction and near-perfect retrieval (>99%) up to at least 10M tokens, a generational leap over existing models such as Claude 3.0 (200k) and GPT-4 Turbo (128k). Finally, we highlight real-world use cases, such as Gemini 1.5 collaborating with professionals on completing their tasks achieving 26 to 75% time savings across 10 different job categories, as well as surprising new capabilities of large language models at the frontier; when given a grammar manual for Kalamang, a language with fewer than 200 speakers worldwide, the model learns to translate English to Kalamang at a similar level to a person who learned from the same content.

Scaling laws are useful guides for derisking expensive training runs, as they predict performance of large models using cheaper, small-scale experiments. However, there remain gaps between current scaling studies and how language models are ultimately trained and evaluated. For instance, scaling is usually studied in the compute-optimal training regime (i.e., "Chinchilla optimal" regime). In contrast, models are often over-trained to reduce inference costs. Moreover, scaling laws mostly predict loss on next-token prediction, but models are usually compared on downstream task performance. To address both shortcomings, we create a testbed of 104 models with 0.011B to 6.9B parameters trained with various numbers of tokens on three data distributions. First, we fit scaling laws that extrapolate in both the amount of over-training and the number of model parameters. This enables us to predict the validation loss of a 1.4B parameter, 900B token run (i.e., 32$\times$ over-trained) and a 6.9B parameter, 138B token run (i.e., a compute-optimal run)$\unicode{x2014}$each from experiments that take 300$\times$ less compute. Second, we relate the perplexity of a language model to its downstream task performance by proposing a power law. We use this law to predict top-1 error averaged over downstream tasks for the two aforementioned models, using experiments that take 20$\times$ less compute. Our experiments are available at https://github.com/mlfoundations/scaling.

Commonly, AI or machine learning (ML) models are evaluated on benchmark datasets. This practice supports innovative methodological research, but benchmark performance can be poorly correlated with performance in real-world applications -- a construct validity issue. To improve the validity and practical usefulness of evaluations, we propose using an estimands framework adapted from international clinical trials guidelines. This framework provides a systematic structure for inference and reporting in evaluations, emphasizing the importance of a well-defined estimation target. We illustrate our proposal on examples of commonly used evaluation methodologies - involving cross-validation, clustering evaluation, and LLM benchmarking - that can lead to incorrect rankings of competing models (rank reversals) with high probability, even when performance differences are large. We demonstrate how the estimands framework can help uncover underlying issues, their causes, and potential solutions. Ultimately, we believe this framework can improve the validity of evaluations through better-aligned inference, and help decision-makers and model users interpret reported results more effectively.