Search

Open community-driven platforms like Chatbot Arena that collect user preference data from site visitors have gained a reputation as one of the most trustworthy publicly available benchmarks for LLM performance. While now standard, it is tricky to implement effective guardrails to collect high-quality annotations from humans. In this paper, we demonstrate that three sources of bad annotations, both malicious and otherwise, can corrupt the reliability of open leaderboard rankings. In particular, we show that only 10\% of poor quality votes by apathetic (site visitors not appropriately incentivized to give correct votes) or adversarial (bad actors seeking to inflate the ranking of a target model) annotators can change the rankings of models by up to 5 places on the leaderboard. Finally, we discuss open challenges in ensuring high-quality human annotations.

With the goal of benchmarking generative systems beyond expert software development ability, we introduce Commit0, a benchmark that challenges AI agents to write libraries from scratch. Agents are provided with a specification document outlining the library's API as well as a suite of interactive unit tests, with the goal of producing an implementation of this API accordingly. The implementation is validated through running these unit tests. As a benchmark, Commit0 is designed to move beyond static one-shot code generation towards agents that must process long-form natural language specifications, adapt to multi-stage feedback, and generate code with complex dependencies. Commit0 also offers an interactive environment where models receive static analysis and execution feedback on the code they generate. Our experiments demonstrate that while current agents can pass some unit tests, none can yet fully reproduce full libraries. Results also show that interactive feedback is quite useful for models to generate code that passes more unit tests, validating the benchmarks that facilitate its use.

Music is a potent form of expression that can communicate, accentuate or even create the emotions of an individual or a collective. Both historically and in contemporary experiences, musical expression was and is commonly instrumentalized for social, political and/or economic purposes. Generative artificial intelligence provides a wealth of both opportunities and challenges with regard to music and its role in society. This paper discusses a proposed project integrating artificial intelligence and popular music, with the ultimate goal of creating a powerful tool for implementing music for social transformation, education, healthcare, and emotional well-being. Given that it is being presented at the outset of a collaboration between a computer scientist/data analyst and an ethnomusicologist/social anthropologist. it is mainly conceptual and somewhat speculative in nature., Comment: Link to the paper:https://aimc2024.pubpub.org/pub/rdulfbve/release/1 Published in The International Conference on AI and Musical Creativity at the University of Oxford (2024) https://aimc2024.pubpub.org/

Data selection can reduce the amount of training data needed to finetune LLMs; however, the efficacy of data selection scales directly with its compute. Motivated by the practical challenge of compute-constrained finetuning, we consider the setting in which both the cost of selecting data and training are budgeted for. We first formalize the problem of data selection with a cost-aware utility function, and model the data selection problem as trading off initial-selection cost for training gain. We run a comprehensive sweep of experiments across multiple tasks, varying compute budget by scaling finetuning tokens, model sizes, and data selection compute. Interestingly we find that many powerful data selection methods are almost never compute-optimal, and that cheaper data selection alternatives dominate both from a theoretical and empirical perspective. For compute-optimal training, we find that perplexity and gradient data selection require training-to-selection model size ratios of 5x and 10x, respectively.

Dense document embeddings are central to neural retrieval. The dominant paradigm is to train and construct embeddings by running encoders directly on individual documents. In this work, we argue that these embeddings, while effective, are implicitly out-of-context for targeted use cases of retrieval, and that a contextualized document embedding should take into account both the document and neighboring documents in context - analogous to contextualized word embeddings. We propose two complementary methods for contextualized document embeddings: first, an alternative contrastive learning objective that explicitly incorporates the document neighbors into the intra-batch contextual loss; second, a new contextual architecture that explicitly encodes neighbor document information into the encoded representation. Results show that both methods achieve better performance than biencoders in several settings, with differences especially pronounced out-of-domain. We achieve state-of-the-art results on the MTEB benchmark with no hard negative mining, score distillation, dataset-specific instructions, intra-GPU example-sharing, or extremely large batch sizes. Our method can be applied to improve performance on any contrastive learning dataset and any biencoder.

We present Bayesian Binary Search (BBS), a novel probabilistic variant of the classical binary search/bisection algorithm. BBS leverages machine learning/statistical techniques to estimate the probability density of the search space and modifies the bisection step to split based on probability density rather than the traditional midpoint, allowing for the learned distribution of the search space to guide the search algorithm. Search space density estimation can flexibly be performed using supervised probabilistic machine learning techniques (e.g., Gaussian process regression, Bayesian neural networks, quantile regression) or unsupervised learning algorithms (e.g., Gaussian mixture models, kernel density estimation (KDE), maximum likelihood estimation (MLE)). We demonstrate significant efficiency gains of using BBS on both simulated data across a variety of distributions and in a real-world binary search use case of probing channel balances in the Bitcoin Lightning Network, for which we have deployed the BBS algorithm in a production setting.

Broad textual understanding and in-context learning require language models that utilize full document contexts. Due to the implementation challenges associated with directly training long-context models, many methods have been proposed for extending models to handle long contexts. However, owing to differences in data and model classes, it has been challenging to compare these approaches, leading to uncertainty as to how to evaluate long-context performance and whether it differs from standard evaluation. We implement a controlled protocol for extension methods with a standardized evaluation, utilizing consistent base models and extension data. Our study yields several insights into long-context behavior. First, we reaffirm the critical role of perplexity as a general-purpose performance indicator even in longer-context tasks. Second, we find that current approximate attention methods systematically underperform across long-context tasks. Finally, we confirm that exact fine-tuning based methods are generally effective within the range of their extension, whereas extrapolation remains challenging. All codebases, models, and checkpoints will be made available open-source, promoting transparency and facilitating further research in this critical area of AI development.

Linear RNN architectures, like Mamba, can be competitive with Transformer models in language modeling while having advantageous deployment characteristics. Given the focus on training large-scale Transformer models, we consider the challenge of converting these pretrained models for deployment. We demonstrate that it is feasible to distill large Transformers into linear RNNs by reusing the linear projection weights from attention layers with academic GPU resources. The resulting hybrid model, which incorporates a quarter of the attention layers, achieves performance comparable to the original Transformer in chat benchmarks and outperforms open-source hybrid Mamba models trained from scratch with trillions of tokens in both chat benchmarks and general benchmarks. Moreover, we introduce a hardware-aware speculative decoding algorithm that accelerates the inference speed of Mamba and hybrid models. Overall we show how, with limited computation resources, we can remove many of the original attention layers and generate from the resulting model more efficiently. Our top-performing model, distilled from Llama3-8B-Instruct, achieves a 29.61 length-controlled win rate on AlpacaEval 2 against GPT-4 and 7.35 on MT-Bench, surpassing the best instruction-tuned linear RNN model., Comment: Code is open-sourced at https://github.com/jxiw/MambaInLlama

$K$-nearest neighbor language models ($k$NN-LMs), which integrate retrieval with next-word prediction, have demonstrated strong performance in language modeling as well as downstream NLP benchmarks. These results have led researchers to argue that models trained on poor quality or outdated data could perform well by employing a $k$NN extension that has access to a higher-quality datastore. In this work, we ask whether this improved ability to recall information really translates into downstream abilities. We extensively evaluate $k$NN-LMs on a diverse set of tasks, ranging from sentiment classification and commonsense reasoning to multi-hop reasoning. Results show that $k$NN-LMs excel at memory-intensive tasks, where utilizing the patterns in the input is sufficient for determining the output, but struggle with reasoning tasks that require integrating multiple pieces of information to derive new knowledge. We further demonstrate through oracle experiments and qualitative analysis that even with perfect retrieval, $k$NN-LMs still fail to determine the correct answers, placing an upper bound on their reasoning performance. Code and datastores are released at https://github.com/GSYfate/knnlm-limits/.

When seeking information from unfamiliar documents, users frequently pose questions that cannot be answered by the documents. While existing large language models (LLMs) identify these unanswerable questions, they do not assist users in reformulating their questions, thereby reducing their overall utility. We curate CouldAsk, an evaluation benchmark composed of existing and new datasets for document-grounded question answering, specifically designed to study reformulating unanswerable questions. We evaluate state-of-the-art open-source and proprietary LLMs on CouldAsk. The results demonstrate the limited capabilities of these models in reformulating questions. Specifically, GPT-4 and Llama2-7B successfully reformulate questions only 26% and 12% of the time, respectively. Error analysis shows that 62% of the unsuccessful reformulations stem from the models merely rephrasing the questions or even generating identical questions. We publicly release the benchmark and the code to reproduce the experiments.

We investigate practical and scalable algorithms for training large language models (LLMs) with user-level differential privacy (DP) in order to provably safeguard all the examples contributed by each user. We study two variants of DP-SGD with: (1) example-level sampling (ELS) and per-example gradient clipping, and (2) user-level sampling (ULS) and per-user gradient clipping. We derive a novel user-level DP accountant that allows us to compute provably tight privacy guarantees for ELS. Using this, we show that while ELS can outperform ULS in specific settings, ULS generally yields better results when each user has a diverse collection of examples. We validate our findings through experiments in synthetic mean estimation and LLM fine-tuning tasks under fixed compute budgets. We find that ULS is significantly better in settings where either (1) strong privacy guarantees are required, or (2) the compute budget is large. Notably, our focus on LLM-compatible training algorithms allows us to scale to models with hundreds of millions of parameters and datasets with hundreds of thousands of users.

To better understand the challenges and opportunities facing community colleges that want to expand apprenticeship opportunities to their students, New America conducted a year-long study. We created an advisory committee to guide this work and spoke to apprenticeship, workforce development, and community college leaders about the community colleges role in expanding apprenticeship. Based on these conversations, we chose case studies and conducted in-depth interviews with leaders from the Community College System of New Hampshire's ApprenticeshipNH, Arapahoe in Colorado, San Jacinto in Texas, and Howard Community College in Maryland's programs in IT and cybersecurity, and Coastal Alabama Community College's nursing apprenticeship. Each of these colleges found a different way to fill the role of intermediary, taking on funding from different sources, using a mix of strategies for sponsorship, and finding place-based approaches to strengthening workforce partnerships. In a variety of sectors, these five colleges found ways to fill the intermediary role that worked for their context, and in doing so provided excellent programming that filled their community's needs. This mosaic of case studies demonstrates that there are some common challenges and successful strategies that colleges looking to serve as apprenticeship intermediaries can learn from.

The high power consumption and latency-sensitive deployments of large language models (LLMs) have motivated efficiency techniques like quantization and sparsity. Contextual sparsity, where the sparsity pattern is input-dependent, is crucial in LLMs because the permanent removal of attention heads or neurons from LLMs can significantly degrade accuracy. Prior work has attempted to model contextual sparsity using neural networks trained to predict activation magnitudes, which can be used to dynamically prune structures with low predicted activation magnitude. In this paper, we look beyond magnitude-based pruning criteria to assess attention head and neuron importance in LLMs. We develop a novel predictor called ShadowLLM, which can shadow the LLM behavior and enforce better sparsity patterns, resulting in over 15% improvement in end-to-end accuracy compared to prior methods. In addition, ShadowLLM achieves up to a 20% speed-up over the state-of-the-art DejaVu framework. These enhancements are validated on Llama-2 and OPT models with up to 30 billion parameters. Our code is available at \href{https://github.com/abdelfattah-lab/shadow_llm/}{ShadowLLM}., Comment: Accepted to EMNLP 2024 (Main, Long Paper)

While diffusion models excel at generating high-quality images, prior work reports a significant performance gap between diffusion and autoregressive (AR) methods in language modeling. In this work, we show that simple masked discrete diffusion is more performant than previously thought. We apply an effective training recipe that improves the performance of masked diffusion models and derive a simplified, Rao-Blackwellized objective that results in additional improvements. Our objective has a simple form -- it is a mixture of classical masked language modeling losses -- and can be used to train encoder-only language models that admit efficient samplers, including ones that can generate arbitrary lengths of text semi-autoregressively like a traditional language model. On language modeling benchmarks, a range of masked diffusion models trained with modern engineering practices achieves a new state-of-the-art among diffusion models, and approaches AR perplexity. We provide the code, along with a blog post and video tutorial on the project page: https://s-sahoo.com/mdlm, Comment: NeurIPS 2024. We provide the code at https://github.com/kuleshov-group/mdlm

Reducing serving cost and latency is a fundamental concern for the deployment of language models (LMs) in business applications. To address this, cascades of LMs offer an effective solution that conditionally employ smaller models for simpler queries. Cascaded systems are typically built with independently trained models, neglecting the advantages of considering inference-time interactions of the cascaded LMs during training. In this paper, we present cascade-aware training(CAT), an approach to optimizing the overall quality-cost performance tradeoff of a cascade of LMs. We achieve inference-time benefits by training the small LM with awareness of its place in a cascade and downstream capabilities. We demonstrate the value of the proposed method with over 60 LM tasks of the SuperGLUE, WMT22, and FLAN2021 datasets., Comment: 22 pages, 13 figures

Optimal transport and the Wasserstein distance $\mathcal{W}_p$ have recently seen a number of applications in the fields of statistics, machine learning, data science, and the physical sciences. These applications are however severely restricted by the curse of dimensionality, meaning that the number of data points needed to estimate these problems accurately increases exponentially in the dimension. To alleviate this problem, a number of variants of $\mathcal{W}_p$ have been introduced. We focus here on one of these variants, namely the max-sliced Wasserstein metric $\overline{\mathcal{W}}_p$. This metric reduces the high-dimensional minimization problem given by $\mathcal{W}_p$ to a maximum of one-dimensional measurements in an effort to overcome the curse of dimensionality. In this note we derive concentration results and upper bounds on the expectation of $\overline{\mathcal{W}}_p$ between the true and empirical measure on unbounded reproducing kernel Hilbert spaces. We show that, under quite generic assumptions, probability measures concentrate uniformly fast in one-dimensional subspaces, at (nearly) parametric rates. Our results rely on an improvement of currently known bounds for $\overline{\mathcal{W}}_p$ in the finite-dimensional case.

There is a well developed renormalization theory of real analytic critical circle maps by de Faria, de Melo, and Yampolsky. In this paper, we extend Yampolsky's result on hyperbolicity of renormalization periodic points to a larger class of dynamical objects, namely critical quasicircle maps, i.e. analytic self homeomorphisms of a quasicircle with a single critical point. Unlike critical circle maps, the inner and outer criticalities of critical quasicircle maps can be distinct. We develop a compact analytic renormalization operator called Corona Renormalization with a hyperbolic fixed point whose stable manifold has codimension one and consists of critical quasicircle maps of the same criticality and periodic type rotation number. Our proof is an adaptation of Pacman Renormalization Theory for Siegel disks as well as rigidity results on the escaping dynamics of transcendental entire functions., Comment: 88 pages, 14 figures. In the new version, there has been a major restructuring to improve the overall presentation, and we have also added a short proof of density of repelling periodic points and no wandering domains for renormalization cascades

Entity disambiguation (ED), which links the mentions of ambiguous entities to their referent entities in a knowledge base, serves as a core component in entity linking (EL). Existing generative approaches demonstrate improved accuracy compared to classification approaches under the standardized ZELDA benchmark. Nevertheless, generative approaches suffer from the need for large-scale pre-training and inefficient generation. Most importantly, entity descriptions, which could contain crucial information to distinguish similar entities from each other, are often overlooked. We propose an encoder-decoder model to disambiguate entities with more detailed entity descriptions. Given text and candidate entities, the encoder learns interactions between the text and each candidate entity, producing representations for each entity candidate. The decoder then fuses the representations of entity candidates together and selects the correct entity. Our experiments, conducted on various entity disambiguation benchmarks, demonstrate the strong and robust performance of this model, particularly +1.5% in the ZELDA benchmark compared with GENRE. Furthermore, we integrate this approach into the retrieval/reader framework and observe +1.5% improvements in end-to-end entity linking in the GERBIL benchmark compared with EntQA., Comment: Accepted at NAACL'24 main

We present DrJAX, a JAX-based library designed to support large-scale distributed and parallel machine learning algorithms that use MapReduce-style operations. DrJAX leverages JAX's sharding mechanisms to enable native targeting of TPUs and state-of-the-art JAX runtimes, including Pathways. DrJAX embeds building blocks for MapReduce computations as primitives in JAX. This enables three key benefits. First, DrJAX computations can be translated directly to XLA HLO, enabling flexible integration with a wide array of ML training platforms. Second, DrJAX computations are fully differentiable. Last, DrJAX computations can be interpreted out to existing batch-processing compute systems, including traditional MapReduce systems like Apache Beam and cross-device compute systems like those powering federated learning applications. We show that DrJAX provides an easily programmable, performant, and scalable framework for parallelized algorithm development. DrJAX is available at \url{https://github.com/google-research/google-research/tree/master/drjax}.

Token-free language models learn directly from raw bytes and remove the inductive bias of subword tokenization. Operating on bytes, however, results in significantly longer sequences. In this setting, standard autoregressive Transformers scale poorly as the effective memory required grows with sequence length. The recent development of the Mamba state space model (SSM) offers an appealing alternative approach with a fixed-sized memory state and efficient decoding. We propose MambaByte, a token-free adaptation of the Mamba SSM trained autoregressively on byte sequences. In terms of modeling, we show MambaByte to be competitive with, and even to outperform, state-of-the-art subword Transformers on language modeling tasks while maintaining the benefits of token-free language models, such as robustness to noise. In terms of efficiency, we develop an adaptation of speculative decoding with tokenized drafting and byte-level verification. This results in a $2.6\times$ inference speedup to the standard MambaByte implementation, showing similar decoding efficiency as the subword Mamba. These findings establish the viability of SSMs in enabling token-free language modeling., Comment: Published at COLM 2024

The Powered by Publics (PxP) initiative has been an ambitious undertaking from the beginning, aiming to produce hundreds of thousands more undergraduate degrees and halving equity gaps for low-income, minoritized, and first-generation students by 2025. APLU has collected student performance data from the 127 participating institutions over the past three years--in 2020, 2021, and 2022--to evaluate progress toward these stretch goals. The purpose of this report is to document the network's progress and share examples of innovation emerging from cross-campus collaborations.

Language models produce a distribution over the next token; can we use this information to recover the prompt tokens? We consider the problem of language model inversion and show that next-token probabilities contain a surprising amount of information about the preceding text. Often we can recover the text in cases where it is hidden from the user, motivating a method for recovering unknown prompts given only the model's current distribution output. We consider a variety of model access scenarios, and show how even without predictions for every token in the vocabulary we can recover the probability vector through search. On Llama-2 7b, our inversion method reconstructs prompts with a BLEU of $59$ and token-level F1 of $78$ and recovers $27\%$ of prompts exactly. Code for reproducing all experiments is available at http://github.com/jxmorris12/vec2text.

Comparative reasoning plays a crucial role in text preference prediction; however, large language models (LLMs) often demonstrate inconsistencies in their reasoning. While approaches like Chain-of-Thought improve accuracy in many other settings, they struggle to consistently distinguish the similarities and differences of complex texts. We introduce SC, a prompting approach that predicts text preferences by generating structured intermediate comparisons. SC begins by proposing aspects of comparison, followed by generating textual comparisons under each aspect. We select consistent comparisons with a pairwise consistency comparator that ensures each aspect's comparisons clearly distinguish differences between texts, significantly reducing hallucination and improving consistency. Our comprehensive evaluations across various NLP tasks, including summarization, retrieval, and automatic rating, demonstrate that SC equips LLMs to achieve state-of-the-art performance in text preference prediction.

As the size of pre-trained speech recognition models increases, running these large models in low-latency or resource-constrained environments becomes challenging. In this work, we leverage pseudo-labelling to assemble a large-scale open-source dataset which we use to distill the Whisper model into a smaller variant, called Distil-Whisper. Using a simple word error rate (WER) heuristic, we select only the highest quality pseudo-labels for training. The distilled model is 5.8 times faster with 51% fewer parameters, while performing to within 1% WER on out-of-distribution test data in a zero-shot transfer setting. Distil-Whisper maintains the robustness of the Whisper model to difficult acoustic conditions, while being less prone to hallucination errors on long-form audio. Distil-Whisper is designed to be paired with Whisper for speculative decoding, yielding a 2 times speed-up while mathematically ensuring the same outputs as the original model. To facilitate further research in this domain, we make our training code, inference code and models publicly accessible., Comment: 30 pages, 2 figures, 25 tables

Photovoltaics (PVs) are a critical technology for curbing growing levels of anthropogenic greenhouse gas emissions, and meeting increases in future demand for low-carbon electricity. In order to fulfil ambitions for net-zero carbon dioxide equivalent (CO2eq) emissions worldwide, the global cumulative capacity of solar PVs must increase by an order of magnitude from 0.9 TWp in 2021 to 8.5 TWp by 2050 according to the International Renewable Energy Agency, which is considered to be a highly conservative estimate. In 2020, the Henry Royce Institute brought together the UK PV community to discuss the critical technological and infrastructure challenges that need to be overcome to address the vast challenges in accelerating PV deployment. Herein, we examine the key developments in the global community, especially the progress made in the field since this earlier roadmap, bringing together experts primarily from the UK across the breadth of the photovoltaics community. The focus is both on the challenges in improving the efficiency, stability and levelized cost of electricity of current technologies for utility-scale PVs, as well as the fundamental questions in novel technologies that can have a significant impact on emerging markets, such as indoor PVs, space PVs, and agrivoltaics. We discuss challenges in advanced metrology and computational tools, as well as the growing synergies between PVs and solar fuels, and offer a perspective on the environmental sustainability of the PV industry. Through this roadmap, we emphasize promising pathways forward in both the short- and long-term, and for communities working on technologies across a range of maturity levels to learn from each other., Comment: 160 pages, 21 figures

Large language models (LLMs) excel at processing and generating both text and code. However, LLMs have had limited applicability in grounded task-oriented dialogue as they are difficult to steer toward task objectives and fail to handle novel grounding. We present a modular and interpretable grounded dialogue system that addresses these shortcomings by composing LLMs with a symbolic planner and grounded code execution. Our system consists of a reader and planner: the reader leverages an LLM to convert partner utterances into executable code, calling functions that perform grounding. The translated code's output is stored to track dialogue state, while a symbolic planner determines the next appropriate response. We evaluate our system's performance on the demanding OneCommon dialogue task, involving collaborative reference resolution on abstract images of scattered dots. Our system substantially outperforms the previous state-of-the-art, including improving task success in human evaluations from 56% to 69% in the most challenging setting., Comment: Accepted to EMNLP 2023

We aim to produce a smaller language model that is aligned to user intent. Previous research has shown that applying distilled supervised fine-tuning (dSFT) on larger models significantly improves task accuracy; however, these models are unaligned, i.e. they do not respond well to natural prompts. To distill this property, we experiment with the use of preference data from AI Feedback (AIF). Starting from a dataset of outputs ranked by a teacher model, we apply distilled direct preference optimization (dDPO) to learn a chat model with significantly improved intent alignment. The approach requires only a few hours of training without any additional sampling during fine-tuning. The final result, Zephyr-7B, sets the state-of-the-art on chat benchmarks for 7B parameter models, and requires no human annotation. In particular, results on MT-Bench show that Zephyr-7B surpasses Llama2-Chat-70B, the best open-access RLHF-based model. Code, models, data, and tutorials for the system are available at https://github.com/huggingface/alignment-handbook.