Long-context processing is a critical ability that constrains the applicability of large language models. Although there exist various methods devoted to enhancing the long-context processing ability of large language models (LLMs), they are developed in an isolated manner and lack systematic analysis and integration of their strengths, hindering further developments. In this paper, we introduce UniMem, a unified framework that reformulates existing long-context methods from the view of memory augmentation of LLMs. UniMem is characterized by four key dimensions: Memory Management, Memory Writing, Memory Reading, and Memory Injection, providing a systematic theory for understanding various long-context methods. We reformulate 16 existing methods based on UniMem and analyze four representative methods: Transformer-XL, Memorizing Transformer, RMT, and Longformer into equivalent UniMem forms to reveal their design principles and strengths. Based on these analyses, we propose UniMix, an innovative approach that integrates the strengths of these algorithms. Experimental results show that UniMix achieves superior performance in handling long contexts with significantly lower perplexity than baselines.
相關內容
The advent of the Transformer architecture has propelled the growth of natural language processing (NLP) models, leading to remarkable achievements in numerous NLP tasks. Yet, the absence of specialized hardware like expansive GPU memory and high-speed interconnects poses challenges for training large-scale models. This makes it daunting for many users to experiment with pre-training and fine-tuning large language models (LLMs). In this study, we introduce \atom, a resilient distributed training framework designed for asynchronous training of vast models in a decentralized setting using cost-effective hardware, including consumer-grade GPUs and Ethernet. Unlike conventional model partitioning methods that distribute sub-models across GPUs, \atom aims to accommodate a complete LLM on one host (peer) through seamlessly model swapping and concurrently trains multiple copies across various peers to optimize training throughput. Through static analysis, \atom identifies the best model partitioning strategy and flawlessly merges model execution with swapping. Key benefits of \atom include: Avoiding the central point of failure found in pipeline parallelism methods. Demonstrating superior performance and scalability compared to closely-integrated pipeline parallelism in slower networks. Our experiments using different GPT-3 model configurations reveal that, in scenarios with suboptimal network connections, \atom can enhance training efficiency up to $20 \times$ when juxtaposed with the state-of-the-art decentralized pipeline parallelism approaches.
Speculative decoding has shown to be an effective method for lossless acceleration of large language models (LLMs) during inference. In each iteration, the algorithm first uses a smaller model to draft a block of tokens. The tokens are then verified by the large model in parallel and only a subset of tokens will be kept to guarantee that the final output follows the distribution of the large model. In all of the prior speculative decoding works, the draft verification is performed token-by-token independently. In this work, we propose a better draft verification algorithm that provides additional wall-clock speedup without incurring additional computation cost and draft tokens. We first formulate the draft verification step as a block-level optimal transport problem. The block-level formulation allows us to consider a wider range of draft verification algorithms and obtain a higher number of accepted tokens in expectation in one draft block. We propose a verification algorithm that achieves the optimal accepted length for the block-level transport problem. We empirically evaluate our proposed block-level verification algorithm in a wide range of tasks and datasets, and observe consistent improvements in wall-clock speedup when compared to token-level verification algorithm. To the best of our knowledge, our work is the first to establish improvement over speculative decoding through a better draft verification algorithm.
Surface parameterization is a fundamental geometry processing problem with rich downstream applications. Traditional approaches are designed to operate on well-behaved mesh models with high-quality triangulations that are laboriously produced by specialized 3D modelers, and thus unable to meet the processing demand for the current explosion of ordinary 3D data. In this paper, we seek to perform UV unwrapping on unstructured 3D point clouds. Technically, we propose ParaPoint, an unsupervised neural learning pipeline for achieving global free-boundary surface parameterization by building point-wise mappings between given 3D points and 2D UV coordinates with adaptively deformed boundaries. We ingeniously construct several geometrically meaningful sub-networks with specific functionalities, and assemble them into a bi-directional cycle mapping framework. We also design effective loss functions and auxiliary differential geometric constraints for the optimization of the neural mapping process. To the best of our knowledge, this work makes the first attempt to investigate neural point cloud parameterization that pursues both global mappings and free boundaries. Experiments demonstrate the effectiveness and inspiring potential of our proposed learning paradigm. The code will be publicly available.
Multimodal large language models (MLLMs) have shown impressive reasoning abilities, which, however, are also more vulnerable to jailbreak attacks than their LLM predecessors. Although still capable of detecting unsafe responses, we observe that safety mechanisms of the pre-aligned LLMs in MLLMs can be easily bypassed due to the introduction of image features. To construct robust MLLMs, we propose ECSO(Eyes Closed, Safety On), a novel training-free protecting approach that exploits the inherent safety awareness of MLLMs, and generates safer responses via adaptively transforming unsafe images into texts to activate intrinsic safety mechanism of pre-aligned LLMs in MLLMs. Experiments on five state-of-the-art (SoTA) MLLMs demonstrate that our ECSO enhances model safety significantly (e.g., a 37.6% improvement on the MM-SafetyBench (SD+OCR), and 71.3% on VLSafe for the LLaVA-1.5-7B), while consistently maintaining utility results on common MLLM benchmarks. Furthermore, we show that ECSO can be used as a data engine to generate supervised-finetuning (SFT) data for MLLM alignment without extra human intervention.
Evaluating the alignment of large language models (LLMs) with user-defined coding preferences is a challenging endeavour that requires assessing intricate textual LLMs' outputs. By relying on automated metrics and static analysis tools, existing benchmarks fail to assess nuances in user instructions and LLM outputs, highlighting the need for large-scale datasets and benchmarks for LLM preference alignment. In this paper, we introduce CodeUltraFeedback, a preference dataset of 10,000 complex instructions to tune and align LLMs to coding preferences through AI feedback. We generate responses to the instructions using a pool of 14 diverse LLMs, which we then annotate according to their alignment with five coding preferences using the LLM-as-a-Judge approach with GPT-3.5, producing both numerical and textual feedback. We also present CODAL-Bench, a benchmark for assessing LLM alignment with these coding preferences. Our results show that CodeLlama-7B-Instruct, aligned through reinforcement learning from AI feedback (RLAIF) with direct preference optimization (DPO) using CodeUltraFeedback's AI feedback data, outperforms 34B LLMs on CODAL-Bench, validating the utility of CodeUltraFeedback for preference tuning. Furthermore, we show our DPO-aligned CodeLlama model improves functional correctness on HumanEval+ compared to the unaligned base model. Therefore, our contributions bridge the gap in preference tuning of LLMs for code and set the stage for further advancements in model alignment and RLAIF for code intelligence. Our code and data are available at //github.com/martin-wey/CodeUltraFeedback.
As large language models (LLMs) take on complex tasks, their inputs are supplemented with longer contexts that incorporate domain knowledge or user-specific information. Yet using long contexts poses a challenge for responsive LLM systems, as nothing can be generated until the whole context is processed by the LLM. While the context-processing delay can be reduced by reusing the KV cache of a context across different inputs, fetching the KV cache, which contains large tensors, over the network can cause extra network delays. CacheGen is a fast context-loading module for LLM systems. First, CacheGen uses a custom tensor encoder, which embraces KV cache's distributional properties, to encode a KV cache into more compact bitstream representations with negligible encoding/decoding overhead. This reduces the bandwidth demand to fetch the KV cache. Second, to maintain low context-loading delay and high generation quality, CacheGen adapts the streaming strategies to cope with changes in available bandwidth. When available bandwidth drops, CacheGen may raise the compression level for a part of the context or choose to recompute its KV cache on the fly. We test CacheGen on four popular LLMs of various sizes and four datasets (662 contexts in total). Compared to the recent systems that reuse the KV cache, CacheGen reduces the KV cache size by 3.7-4.3x and the total delay in fetching and processing contexts by 2.7-3.2x while having negligible impact on the LLM response quality in accuracy or perplexity.
Large language models (LLMs) have demonstrated impressive capabilities in natural language processing. However, their internal mechanisms are still unclear and this lack of transparency poses unwanted risks for downstream applications. Therefore, understanding and explaining these models is crucial for elucidating their behaviors, limitations, and social impacts. In this paper, we introduce a taxonomy of explainability techniques and provide a structured overview of methods for explaining Transformer-based language models. We categorize techniques based on the training paradigms of LLMs: traditional fine-tuning-based paradigm and prompting-based paradigm. For each paradigm, we summarize the goals and dominant approaches for generating local explanations of individual predictions and global explanations of overall model knowledge. We also discuss metrics for evaluating generated explanations, and discuss how explanations can be leveraged to debug models and improve performance. Lastly, we examine key challenges and emerging opportunities for explanation techniques in the era of LLMs in comparison to conventional machine learning models.
The emergence of large language models (LLMs) has substantially influenced natural language processing, demonstrating exceptional results across various tasks. In this study, we employ ``Introspective Tips" to facilitate LLMs in self-optimizing their decision-making. By introspectively examining trajectories, LLM refines its policy by generating succinct and valuable tips. Our method enhances the agent's performance in both few-shot and zero-shot learning situations by considering three essential scenarios: learning from the agent's past experiences, integrating expert demonstrations, and generalizing across diverse games. Importantly, we accomplish these improvements without fine-tuning the LLM parameters; rather, we adjust the prompt to generalize insights from the three aforementioned situations. Our framework not only supports but also emphasizes the advantage of employing LLM in in-contxt decision-making. Experiments involving over 100 games in TextWorld illustrate the superior performance of our approach.
Pre-trained language representation models, such as BERT, capture a general language representation from large-scale corpora, but lack domain-specific knowledge. When reading a domain text, experts make inferences with relevant knowledge. For machines to achieve this capability, we propose a knowledge-enabled language representation model (K-BERT) with knowledge graphs (KGs), in which triples are injected into the sentences as domain knowledge. However, too much knowledge incorporation may divert the sentence from its correct meaning, which is called knowledge noise (KN) issue. To overcome KN, K-BERT introduces soft-position and visible matrix to limit the impact of knowledge. K-BERT can easily inject domain knowledge into the models by equipped with a KG without pre-training by-self because it is capable of loading model parameters from the pre-trained BERT. Our investigation reveals promising results in twelve NLP tasks. Especially in domain-specific tasks (including finance, law, and medicine), K-BERT significantly outperforms BERT, which demonstrates that K-BERT is an excellent choice for solving the knowledge-driven problems that require experts.
With the capability of modeling bidirectional contexts, denoising autoencoding based pretraining like BERT achieves better performance than pretraining approaches based on autoregressive language modeling. However, relying on corrupting the input with masks, BERT neglects dependency between the masked positions and suffers from a pretrain-finetune discrepancy. In light of these pros and cons, we propose XLNet, a generalized autoregressive pretraining method that (1) enables learning bidirectional contexts by maximizing the expected likelihood over all permutations of the factorization order and (2) overcomes the limitations of BERT thanks to its autoregressive formulation. Furthermore, XLNet integrates ideas from Transformer-XL, the state-of-the-art autoregressive model, into pretraining. Empirically, XLNet outperforms BERT on 20 tasks, often by a large margin, and achieves state-of-the-art results on 18 tasks including question answering, natural language inference, sentiment analysis, and document ranking.
小貼士
登錄享
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191