PyTorchKR Papers

최신 논문

총 486편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2407.17406공식 코드 · Papers with Code

Dependency Transformer Grammars: Integrating Dependency Structures into Transformer Language Models

Syntactic Transformer language models aim to achieve better generalization through simultaneously modeling syntax trees and sentences. While prior work has been focusing on adding constituency-based structures to Transformers, we introduce Dependency Transformer Grammars (DTGs), a new class of Transformer language model with explicit dependency-based inductive bias. DTGs simulate dependency transition systems with constrained attention patterns by modifying attention masks, incorporate the stack information through relative positional encoding, and augment dependency arc representation with a combination of token embeddings and operation embeddings. When trained on a dataset of sentences annotated with dependency trees, DTGs achieve better generalization while maintaining comparable perplexity with Transformer language model baselines. DTGs also outperform recent constituency-based models, showing that dependency can better guide Transformer language models. Our code is released at https://github.com/zhaoyd1/Dep_Transformer_Grammars.

arXiv:2407.12844공식 코드 · Papers with Code

$\texttt{metabench}$ -- A Sparse Benchmark to Measure General Ability in Large Language Models

Large Language Models (LLMs) vary in their abilities on a range of tasks. Initiatives such as the $\texttt{Open LLM Leaderboard}$ aim to quantify these differences with several large benchmarks (sets of test items to which an LLM can respond either correctly or incorrectly). However, high correlations within and between benchmark scores suggest that (1) there exists a small set of common underlying abilities that these benchmarks measure, and (2) items tap into redundant information and the benchmarks may thus be considerably compressed. We use data from $n > 5000$ LLMs to identify the most informative items of six benchmarks, ARC, GSM8K, HellaSwag, MMLU, TruthfulQA and WinoGrande (with $d=28,632$ items in total). From them we distill a sparse benchmark, $\texttt{metabench}$, that has less than $3\%$ of the original size of all six benchmarks combined. This new sparse benchmark goes beyond point scores by yielding estimators of the underlying benchmark-specific abilities. We show that these estimators (1) can be used to reconstruct each original $\textit{individual}$ benchmark score with, on average, $1.5\%$ root mean square error (RMSE), (2) reconstruct the original $\textit{total}$ score with $0.8\%$ RMSE, and (3) have a single underlying common factor whose Spearman correlation with the total score is $r = 0.93$.

arXiv:2407.13623공식 코드 · Papers with Code

Scaling Laws with Vocabulary: Larger Models Deserve Larger Vocabularies

Research on scaling large language models (LLMs) has primarily focused on model parameters and training data size, overlooking the role of vocabulary size. We investigate how vocabulary size impacts LLM scaling laws by training models ranging from 33M to 3B parameters on up to 500B characters with various vocabulary configurations. We propose three complementary approaches for predicting the compute-optimal vocabulary size: IsoFLOPs analysis, derivative estimation, and parametric fit of the loss function. Our approaches converge on the conclusion that the optimal vocabulary size depends on the compute budget, with larger models requiring larger vocabularies. Most LLMs, however, use insufficient vocabulary sizes. For example, we predict that the optimal vocabulary size of Llama2-70B should have been at least 216K, 7 times larger than its vocabulary of 32K. We validate our predictions empirically by training models with 3B parameters across different FLOPs budgets. Adopting our predicted optimal vocabulary size consistently improves downstream performance over commonly used vocabulary sizes. By increasing the vocabulary size from the conventional 32K to 43K, we improve performance on ARC-Challenge from 29.1 to 32.0 with the same 2.3e21 FLOPs. Our work highlights the importance of jointly considering tokenization and model scaling for efficient pre-training. The code and demo are available at https://github.com/sail-sg/scaling-with-vocab and https://hf.co/spaces/sail/scaling-with-vocab-demo.

arXiv:2407.08812

Fence decompositions and cherry covers in non-binary phylogenetic networks

Reticulate evolution can be modelled using phylogenetic networks. Tree-based networks, which are one of the more general classes of phylogenetic networks, have recently gained eminence for its ability to represent evolutionary histories with an underlying tree structure. To better understand tree-based networks, numerous characterizations have been proposed, based on tree embeddings, matchings, and arc partitions. Here, we build a bridge between two arc partition characterizations, namely maximal fence decompositions and cherry covers. Results on cherry covers have been found for general phylogenetic networks. We first show that the number of cherry covers is the same as the number of support trees (underlying tree structure of tree-based networks) for a given semibinary network. Maximal fence decompositions have only been defined thus far for binary networks (constraints on vertex degrees). We remedy this by generalizing fence decompositions to non-binary networks, and using this, we characterize semi-binary tree-based networks in terms of forbidden structures. Furthermore, we give an explicit enumeration of cherry covers of semi-binary networks, by studying its fence decomposition. Finally, we prove that it is possible to characterize semi-binary tree-child networks, a subclass of tree-based networks, in terms of the number of their cherry covers.

arXiv:2407.09127

Robustness of Explainable Artificial Intelligence in Industrial Process Modelling

eXplainable Artificial Intelligence (XAI) aims at providing understandable explanations of black box models. In this paper, we evaluate current XAI methods by scoring them based on ground truth simulations and sensitivity analysis. To this end, we used an Electric Arc Furnace (EAF) model to better understand the limits and robustness characteristics of XAI methods such as SHapley Additive exPlanations (SHAP), Local Interpretable Model-agnostic Explanations (LIME), as well as Averaged Local Effects (ALE) or Smooth Gradients (SG) in a highly topical setting. These XAI methods were applied to various types of black-box models and then scored based on their correctness compared to the ground-truth sensitivity of the data-generating processes using a novel scoring evaluation methodology over a range of simulated additive noise. The resulting evaluation shows that the capability of the Machine Learning (ML) models to capture the process accurately is, indeed, coupled with the correctness of the explainability of the underlying data-generating process. We furthermore show the differences between XAI methods in their ability to correctly predict the true sensitivity of the modeled industrial process.

arXiv:2407.08223

Speculative RAG: Enhancing Retrieval Augmented Generation through Drafting

Retrieval augmented generation (RAG) combines the generative abilities of large language models (LLMs) with external knowledge sources to provide more accurate and up-to-date responses. Recent RAG advancements focus on improving retrieval outcomes through iterative LLM refinement or self-critique capabilities acquired through additional instruction tuning of LLMs. In this work, we introduce Speculative RAG - a framework that leverages a larger generalist LM to efficiently verify multiple RAG drafts produced in parallel by a smaller, distilled specialist LM. Each draft is generated from a distinct subset of retrieved documents, offering diverse perspectives on the evidence while reducing input token counts per draft. This approach enhances comprehension of each subset and mitigates potential position bias over long context. Our method accelerates RAG by delegating drafting to the smaller specialist LM, with the larger generalist LM performing a single verification pass over the drafts. Extensive experiments demonstrate that Speculative RAG achieves state-of-the-art performance with reduced latency on TriviaQA, MuSiQue, PubHealth, and ARC-Challenge benchmarks. It notably enhances accuracy by up to 12.97% while reducing latency by 51% compared to conventional RAG systems on PubHealth.

arXiv:2406.17681공식 코드 · Papers with Code

VarBench: Robust Language Model Benchmarking Through Dynamic Variable Perturbation

As large language models achieve impressive scores on traditional benchmarks, an increasing number of researchers are becoming concerned about benchmark data leakage during pre-training, commonly known as the data contamination problem. To ensure fair evaluation, recent benchmarks release only the training and validation sets, keeping the test set labels closed-source. They require anyone wishing to evaluate his language model to submit the model's predictions for centralized processing and then publish the model's result on their leaderboard. However, this submission process is inefficient and prevents effective error analysis. To address this issue, we propose to variabilize benchmarks and evaluate language models dynamically. Specifically, we extract variables from each test case and define a value range for each variable. For each evaluation, we sample new values from these value ranges to create unique test cases, thus ensuring a fresh evaluation each time. We applied this variable perturbation method to four datasets: GSM8K, ARC, CommonsenseQA, and TruthfulQA, which cover mathematical generation and multiple-choice tasks. Our experimental results demonstrate that this approach provides a more accurate assessment of the true capabilities of language models, effectively mitigating the contamination problem.

arXiv:2406.17557

The FineWeb Datasets: Decanting the Web for the Finest Text Data at Scale

The performance of a large language model (LLM) depends heavily on the quality and size of its pretraining dataset. However, the pretraining datasets for state-of-the-art open LLMs like Llama 3 and Mixtral are not publicly available and very little is known about how they were created. In this work, we introduce FineWeb, a 15-trillion token dataset derived from 96 Common Crawl snapshots that produces better-performing LLMs than other open pretraining datasets. To advance the understanding of how best to curate high-quality pretraining datasets, we carefully document and ablate all of the design choices used in FineWeb, including in-depth investigations of deduplication and filtering strategies. In addition, we introduce FineWeb-Edu, a 1.3-trillion token collection of educational text filtered from FineWeb. LLMs pretrained on FineWeb-Edu exhibit dramatically better performance on knowledge- and reasoning-intensive benchmarks like MMLU and ARC. Along with our datasets, we publicly release our data curation codebase and all of the models trained during our ablation experiments.

arXiv:2406.17663

LLM-ARC: Enhancing LLMs with an Automated Reasoning Critic

We introduce LLM-ARC, a neuro-symbolic framework designed to enhance the logical reasoning capabilities of Large Language Models (LLMs), by combining them with an Automated Reasoning Critic (ARC). LLM-ARC employs an Actor-Critic method where the LLM Actor generates declarative logic programs along with tests for semantic correctness, while the Automated Reasoning Critic evaluates the code, runs the tests and provides feedback on test failures for iterative refinement. Implemented using Answer Set Programming (ASP), LLM-ARC achieves a new state-of-the-art accuracy of 88.32% on the FOLIO benchmark which tests complex logical reasoning capabilities. Our experiments demonstrate significant improvements over LLM-only baselines, highlighting the importance of logic test generation and iterative self-refinement. We achieve our best result using a fully automated self-supervised training loop where the Actor is trained on end-to-end dialog traces with Critic feedback. We discuss potential enhancements and provide a detailed error analysis, showcasing the robustness and efficacy of LLM-ARC for complex natural language reasoning tasks.

arXiv:2406.16061

PORT: Preference Optimization on Reasoning Traces

Preference optimization methods have been successfully applied to improve not only the alignment of large language models (LLMs) with human values, but also specific natural language tasks such as summarization and stylistic continuations. This paper proposes using preference optimization methods on Chain-of-Thought steps in order to improve the reasoning performances of language models. While the chosen answers are obtained from datasets that include reasoning traces, we propose two complementary schemes for generating rejected answers: digit corruption, and weak LLM prompting. Our approach leads to increased accuracy on the GSM8K, AQuA-RAT, and ARC benchmarks for Falcon2-11B and Mistral-7B. For example, the approach can lead to up to a relative 8.47% increase in accuracy on the GSM8K benchmark without any extra annotations. This work suggests that spending resources on creating more datasets of reasoning traces would further boost LLM performances on informal reasoning tasks.

arXiv:2406.13233공식 코드 · Papers with Code

AdaMoE: Token-Adaptive Routing with Null Experts for Mixture-of-Experts Language Models

Mixture of experts (MoE) has become the standard for constructing production-level large language models (LLMs) due to its promise to boost model capacity without causing significant overheads. Nevertheless, existing MoE methods usually enforce a constant top-k routing for all tokens, which is arguably restrictive because various tokens (e.g., "<EOS>" vs. "apple") may require various numbers of experts for feature abstraction. Lifting such a constraint can help make the most of limited resources and unleash the potential of the model for downstream tasks. In this sense, we introduce AdaMoE to realize token-adaptive routing for MoE, where different tokens are permitted to select a various number of experts. AdaMoE makes minimal modifications to the vanilla MoE with top-k routing -- it simply introduces a fixed number of null experts, which do not consume any FLOPs, to the expert set and increases the value of k. AdaMoE does not force each token to occupy a fixed number of null experts but ensures the average usage of the null experts with a load-balancing loss, leading to an adaptive number of null/true experts used by each token. AdaMoE exhibits a strong resemblance to MoEs with expert choice routing while allowing for trivial auto-regressive modeling. AdaMoE is easy to implement and can be effectively applied to pre-trained (MoE-)LLMs. Extensive studies show that AdaMoE can reduce average expert load (FLOPs) while achieving superior performance. For example, on the ARC-C dataset, applying our method to fine-tuning Mixtral-8x7B can reduce FLOPs by 14.5% while increasing accuracy by 1.69%.

arXiv:2406.12098

Circular transformation of the European steel industry renders scrap metal a strategic resource

The steel industry is a major contributor to CO2 emissions, accounting for 7% of global emissions. The European steel industry is seeking to reduce its emissions by increasing the use of electric arc furnaces (EAFs), which can produce steel from scrap, marking a major shift towards a circular steel economy. Here, we show by combining trade with business intelligence data that this shift requires a deep restructuring of the global and European scrap trade, as well as a substantial scaling of the underlying business ecosystem. We find that the scrap imports of European countries with major EAF installations have steadily decreased since 2007 while globally scrap trade started to increase recently. Our statistical modelling shows that every 1,000 tonnes of EAF capacity installed is associated with an increase in annual imports of 550 tonnes and a decrease in annual exports of 1,000 tonnes of scrap, suggesting increased competition for scrap metal as countries ramp up their EAF capacity. Furthermore, each scrap company enables an increase of around 79,000 tonnes of EAF-based steel production per year in the EU. Taking these relations as causal and extrapolating to the currently planned EAF capacity, we find that an additional 730 (SD 140) companies might be required, employing about 35,000 people (IQR 29,000-50,000) and generating an additional estimated turnover of USD 35 billion (IQR 27-48). Our results thus suggest that scrap metal is likely to become a strategic resource. They highlight the need for a massive restructuring of the industry's supply networks and identify the resulting growth opportunities for companies.

arXiv:2406.11473

Promises, Outlooks and Challenges of Diffusion Language Modeling

The modern autoregressive Large Language Models (LLMs) have achieved outstanding performance on NLP benchmarks, and they are deployed in the real world. However, they still suffer from limitations of the autoregressive training paradigm. For example, autoregressive token generation is notably slow and can be prone to \textit{exposure bias}. The diffusion-based language models were proposed as an alternative to autoregressive generation to address some of these limitations. We evaluate the recently proposed Score Entropy Discrete Diffusion (SEDD) approach and show it is a promising alternative to autoregressive generation but it has some short-comings too. We empirically demonstrate the advantages and challenges of SEDD, and observe that SEDD generally matches autoregressive models in perplexity and on benchmarks such as HellaSwag, Arc or WinoGrande. Additionally, we show that in terms of inference latency, SEDD can be up to 4.5$\times$ more efficient than GPT-2. While SEDD allows conditioning on tokens at abitrary positions, SEDD appears slightly weaker than GPT-2 for conditional generation given short prompts. Finally, we reproduced the main results from the original SEDD paper.

arXiv:2406.07644

Regularizing Numerical Extremals Along Singular Arcs: A Lie-Theoretic Approach

Numerical ``direct'' approaches to time-optimal control often fail to find solutions that are singular in the sense of the Pontryagin Maximum Principle, performing better when searching for saturated (bang-bang) solutions. In previous work by one of the authors, singular solutions were shown to exist for the time-optimal control problem for fully actuated mechanical systems under hard torque constraints. Explicit formulas, based on a Lie theoretic analysis of the problem, were given for singular segments of trajectories, but the global structure of solutions remains unknown. In this work, we review the aforementioned framework, and show how to effectively combine these formulas with the use of general-purpose optimal control software packages. By using the explicit formula given by the theory in the intervals where the numerical solution enters a singular arc, we not only obtain an algebraic expression for the control in that interval but we are also able to remove artifacts present in the numerical solution. In this way, the best features of numerical algorithms and theory complement each other and provide a better picture of the global optimal structure. We illustrate the technique on a two degree of freedom robotic arm example, using two distinct optimal control numerical software packages running on different programming languages.

arXiv:2406.09016공식 코드 · Papers with Code

Cross-Modal Learning for Anomaly Detection in Complex Industrial Process: Methodology and Benchmark

Anomaly detection in complex industrial processes plays a pivotal role in ensuring efficient, stable, and secure operation. Existing anomaly detection methods primarily focus on analyzing dominant anomalies using the process variables (such as arc current) or constructing neural networks based on abnormal visual features, while overlooking the intrinsic correlation of cross-modal information. This paper proposes a cross-modal Transformer (dubbed FmFormer), designed to facilitate anomaly detection by exploring the correlation between visual features (video) and process variables (current) in the context of the fused magnesium smelting process. Our approach introduces a novel tokenization paradigm to effectively bridge the substantial dimensionality gap between the 3D video modality and the 1D current modality in a multiscale manner, enabling a hierarchical reconstruction of pixel-level anomaly detection. Subsequently, the FmFormer leverages self-attention to learn internal features within each modality and bidirectional cross-attention to capture correlations across modalities. By decoding the bidirectional correlation features, we obtain the final detection result and even locate the specific anomaly region. To validate the effectiveness of the proposed method, we also present a pioneering cross-modal benchmark of the fused magnesium smelting process, featuring synchronously acquired video and current data for over 2.2 million samples. Leveraging cross-modal learning, the proposed FmFormer achieves state-of-the-art performance in detecting anomalies, particularly under extreme interferences such as current fluctuations and visual occlusion caused by heavy water mist. The presented methodology and benchmark may be applicable to other industrial applications with some amendments. The benchmark will be released at https://github.com/GaochangWu/FMF-Benchmark.

arXiv:2406.05673공식 코드 · Papers with Code

Flow of Reasoning:Training LLMs for Divergent Problem Solving with Minimal Examples

The ability to generate diverse solutions to a given problem is a hallmark of human creativity. This divergent reasoning is also crucial for machines, enhancing their robustness and enabling them to assist humans in many applications such as scientific discovery. However, existing approaches to multi-step reasoning with large language models (LLMs) have mostly focused only on reasoning accuracy, without further discovering more diverse valid solutions. For example, supervised fine-tuning can improve LLM reasoning quality, but requires extensive supervised data to capture the full range of possible solutions. Reinforcement learning aims to find limited highest-reward solutions while neglecting the solution diversity. To fill this gap, we propose Flow of Reasoning (FoR), an efficient diversity-seeking LLM finetuning method aimed at improving reasoning quality and diversity with minimal data. FoR formulates multi-step LLM reasoning as a Markovian flow on a DAG-structured reasoning graph. This formulation allows us to incorporate and adapt principled GFlowNet approaches, for finetuning LLMs to sample diverse reasoning paths with probabilities proportional to the (unnormalized) reward of target problems. Extensive experiments show that, with limited training examples (e.g., 15 examples), FoR enables the discovery of diverse, creative, high-quality solutions, greatly outperforming a wide range of existing inference and training methods across five challenging puzzle-solving tasks, including BlocksWorld (embodied reasoning), Game24 (math puzzle solving), Rubik's Cube (spatial reasoning), 1D-ARC (abstraction reasoning), and PrOntoQA (logical reasoning). Code is available at https://github.com/Yu-Fangxu/FoR.

arXiv:2405.17976공식 코드 · Papers with Code

Yuan 2.0-M32: Mixture of Experts with Attention Router

Yuan 2.0-M32, with a similar base architecture as Yuan-2.0 2B, uses a mixture-of-experts architecture with 32 experts of which 2 experts are active. A new router network, Attention Router, is proposed and adopted for a more efficient selection of experts, which improves the accuracy compared to the model with classical router network. Yuan 2.0-M32 is trained with 2000B tokens from scratch, and the training computation consumption is only 9.25% of a dense model at the same parameter scale. Yuan 2.0-M32 demonstrates competitive capability on coding, math, and various domains of expertise, with only 3.7B active parameters of 40B in total, and 7.4 GFlops forward computation per token, both of which are only 1/19 of Llama3-70B. Yuan 2.0-M32 surpass Llama3-70B on MATH and ARC-Challenge benchmark, with accuracy of 55.89 and 95.8 respectively. The models and source codes of Yuan 2.0-M32 are released at Github1.

arXiv:2405.17361

A One-Layer Decoder-Only Transformer is a Two-Layer RNN: With an Application to Certified Robustness

This paper reveals a key insight that a one-layer decoder-only Transformer is equivalent to a two-layer Recurrent Neural Network (RNN). Building on this insight, we propose ARC-Tran, a novel approach for verifying the robustness of decoder-only Transformers against arbitrary perturbation spaces. Compared to ARC-Tran, current robustness verification techniques are limited either to specific and length-preserving perturbations like word substitutions or to recursive models like LSTMs. ARC-Tran addresses these limitations by meticulously managing position encoding to prevent mismatches and by utilizing our key insight to achieve precise and scalable verification. Our evaluation shows that ARC-Tran (1) trains models more robust to arbitrary perturbation spaces than those produced by existing techniques and (2) shows high certification accuracy of the resulting models.

arXiv:2405.16771공식 코드 · Papers with Code

ARC: A Generalist Graph Anomaly Detector with In-Context Learning

Graph anomaly detection (GAD), which aims to identify abnormal nodes that differ from the majority within a graph, has garnered significant attention. However, current GAD methods necessitate training specific to each dataset, resulting in high training costs, substantial data requirements, and limited generalizability when being applied to new datasets and domains. To address these limitations, this paper proposes ARC, a generalist GAD approach that enables a ``one-for-all'' GAD model to detect anomalies across various graph datasets on-the-fly. Equipped with in-context learning, ARC can directly extract dataset-specific patterns from the target dataset using few-shot normal samples at the inference stage, without the need for retraining or fine-tuning on the target dataset. ARC comprises three components that are well-crafted for capturing universal graph anomaly patterns: 1) smoothness-based feature Alignment module that unifies the features of different datasets into a common and anomaly-sensitive space; 2) ego-neighbor Residual graph encoder that learns abnormality-related node embeddings; and 3) cross-attentive in-Context anomaly scoring module that predicts node abnormality by leveraging few-shot normal samples. Extensive experiments on multiple benchmark datasets from various domains demonstrate the superior anomaly detection performance, efficiency, and generalizability of ARC.

arXiv:2405.14432

Adaptive Gradient Clipping for Robust Federated Learning

Robust federated learning aims to maintain reliable performance despite the presence of adversarial or misbehaving workers. While state-of-the-art (SOTA) robust distributed gradient descent (Robust-DGD) methods were proven theoretically optimal, their empirical success has often relied on pre-aggregation gradient clipping. However, existing static clipping strategies yield inconsistent results: enhancing robustness against some attacks while being ineffective or even detrimental against others. To address this limitation, we propose a principled adaptive clipping strategy, Adaptive Robust Clipping (ARC), which dynamically adjusts clipping thresholds based on the input gradients. We prove that ARC not only preserves the theoretical robustness guarantees of SOTA Robust-DGD methods but also provably improves asymptotic convergence when the model is well-initialized. Extensive experiments on benchmark image classification tasks confirm these theoretical insights, demonstrating that ARC significantly enhances robustness, particularly in highly heterogeneous and adversarial settings.