This work studies post-training parameter quantization in large language models (LLMs). We introduce quantization with incoherence processing (QuIP), a new method based on the insight that quantization benefits from incoherent weight and Hessian matrices, i.e., from the weights and the directions in which it is important to round them accurately being unaligned with the coordinate axes. QuIP consists of two steps: (1) an adaptive rounding procedure minimizing a quadratic proxy objective; (2) efficient pre- and post-processing that ensures weight and Hessian incoherence via multiplication by random orthogonal matrices. We complement QuIP with the first theoretical analysis for an LLM-scale quantization algorithm, and show that our theory also applies to an existing method, OPTQ. Empirically, we find that our incoherence preprocessing improves several existing quantization algorithms and yields the first LLM quantization methods that produce viable results using only two bits per weight. Our code can be found at //github.com/jerry-chee/QuIP .
相關內容
As large language models (LLM) evolve in their capabilities, various recent studies have tried to quantify their behavior using psychological tools created to study human behavior. One such example is the measurement of "personality" of LLMs using personality self-assessment tests. In this paper, we take three such studies on personality measurement of LLMs that use personality self-assessment tests created to study human behavior. We use the prompts used in these three different papers to measure the personality of the same LLM. We find that all three prompts lead very different personality scores. This simple test reveals that personality self-assessment scores in LLMs depend on the subjective choice of the prompter. Since we don't know the ground truth value of personality scores for LLMs as there is no correct answer to such questions, there's no way of claiming if one prompt is more or less correct than the other. We then introduce the property of option order symmetry for personality measurement of LLMs. Since most of the self-assessment tests exist in the form of multiple choice question (MCQ) questions, we argue that the scores should also be robust to not just the prompt template but also the order in which the options are presented. This test unsurprisingly reveals that the answers to the self-assessment tests are not robust to the order of the options. These simple tests, done on ChatGPT and Llama2 models show that self-assessment personality tests created for humans are not appropriate for measuring personality in LLMs.
Oobleck enables resilient distributed training of large DNN models with guaranteed fault tolerance. It takes a planning-execution co-design approach, where it first generates a set of heterogeneous pipeline templates and instantiates at least $f+1$ logically equivalent pipeline replicas to tolerate any $f$ simultaneous failures. During execution, it relies on already-replicated model states across the replicas to provide fast recovery. Oobleck provably guarantees that some combination of the initially created pipeline templates can be used to cover all available resources after $f$ or fewer simultaneous failures, thereby avoiding resource idling at all times. Evaluation on large DNN models with billions of parameters shows that Oobleck provides consistently high throughput, and it outperforms state-of-the-art fault tolerance solutions like Bamboo and Varuna by up to $13.9x$.
Large language models (LLMs) have shown great promise for capturing contextual information in natural language processing tasks. We propose a novel approach to speaker diarization that incorporates the prowess of LLMs to exploit contextual cues in human dialogues. Our method builds upon an acoustic-based speaker diarization system by adding lexical information from an LLM in the inference stage. We model the multi-modal decoding process probabilistically and perform joint acoustic and lexical beam search to incorporate cues from both modalities: audio and text. Our experiments demonstrate that infusing lexical knowledge from the LLM into an acoustics-only diarization system improves overall speaker-attributed word error rate (SA-WER). The experimental results show that LLMs can provide complementary information to acoustic models for the speaker diarization task via proposed beam search decoding approach showing up to 39.8% relative delta-SA-WER improvement from the baseline system. Thus, we substantiate that the proposed technique is able to exploit contextual information that is inaccessible to acoustics-only systems which is represented by speaker embeddings. In addition, these findings point to the potential of using LLMs to improve speaker diarization and other speech processing tasks by capturing semantic and contextual cues.
Large language models (LLMs) often demonstrate inconsistencies with human preferences. Previous research gathered human preference data and then aligned the pre-trained models using reinforcement learning or instruction tuning, the so-called finetuning step. In contrast, aligning frozen LLMs without any extra data is more appealing. This work explores the potential of the latter setting. We discover that by integrating self-evaluation and rewind mechanisms, unaligned LLMs can directly produce responses consistent with human preferences via self-boosting. We introduce a novel inference method, Rewindable Auto-regressive INference (RAIN), that allows pre-trained LLMs to evaluate their own generation and use the evaluation results to guide backward rewind and forward generation for AI safety. Notably, RAIN operates without the need of extra data for model alignment and abstains from any training, gradient computation, or parameter updates; during the self-evaluation phase, the model receives guidance on which human preference to align with through a fixed-template prompt, eliminating the need to modify the initial prompt. Experimental results evaluated by GPT-4 and humans demonstrate the effectiveness of RAIN: on the HH dataset, RAIN improves the harmlessness rate of LLaMA 30B over vanilla inference from 82% to 97%, while maintaining the helpfulness rate. Under the leading adversarial attack llm-attacks on Vicuna 33B, RAIN establishes a new defense baseline by reducing the attack success rate from 94% to 19%.
Multi-modal large language models (MLLMs) are trained based on large language models (LLM), with an enhanced capability to comprehend multi-modal inputs and generate textual responses. While they excel in multi-modal tasks, the pure NLP abilities of MLLMs are often underestimated and left untested. In this study, we get out of the box and unveil an intriguing characteristic of MLLMs -- our preliminary results suggest that visual instruction tuning, a prevailing strategy for transitioning LLMs into MLLMs, unexpectedly and interestingly helps models attain both improved truthfulness and ethical alignment in the pure NLP context. For example, a visual-instruction-tuned LLaMA2 7B model surpasses the performance of the LLaMA2-chat 7B model, fine-tuned with over one million human annotations, on TruthfulQA-mc and Ethics benchmarks. Further analysis reveals that the improved alignment can be attributed to the superior instruction quality inherent to visual-text data. In releasing our code at github.com/UCSC-VLAA/Sight-Beyond-Text, we aspire to foster further exploration into the intrinsic value of visual-text synergies and, in a broader scope, multi-modal interactions in alignment research.
Deep neural networks (DNNs) underpin many machine learning applications. Production quality DNN models achieve high inference accuracy by training millions of DNN parameters which has a significant resource footprint. This presents a challenge for resources operating at the extreme edge of the network, such as mobile and embedded devices that have limited computational and memory resources. To address this, models are pruned to create lightweight, more suitable variants for these devices. Existing pruning methods are unable to provide similar quality models compared to their unpruned counterparts without significant time costs and overheads or are limited to offline use cases. Our work rapidly derives suitable model variants while maintaining the accuracy of the original model. The model variants can be swapped quickly when system and network conditions change to match workload demand. This paper presents DNNShifter, an end-to-end DNN training, spatial pruning, and model switching system that addresses the challenges mentioned above. At the heart of DNNShifter is a novel methodology that prunes sparse models using structured pruning. The pruned model variants generated by DNNShifter are smaller in size and thus faster than dense and sparse model predecessors, making them suitable for inference at the edge while retaining near similar accuracy as of the original dense model. DNNShifter generates a portfolio of model variants that can be swiftly interchanged depending on operational conditions. DNNShifter produces pruned model variants up to 93x faster than conventional training methods. Compared to sparse models, the pruned model variants are up to 5.14x smaller and have a 1.67x inference latency speedup, with no compromise to sparse model accuracy. In addition, DNNShifter has up to 11.9x lower overhead for switching models and up to 3.8x lower memory utilisation than existing approaches.
We introduce RL4CO, an extensive reinforcement learning (RL) for combinatorial optimization (CO) benchmark. RL4CO employs state-of-the-art software libraries as well as best practices in implementation, such as modularity and configuration management, to be efficient and easily modifiable by researchers for adaptations of neural network architecture, environments, and RL algorithms. Contrary to the existing focus on specific tasks like the traveling salesman problem (TSP) for performance assessment, we underline the importance of scalability and generalization capabilities for diverse CO tasks. We also systematically benchmark zero-shot generalization, sample efficiency, and adaptability to changes in data distributions of various models. Our experiments show that some recent SOTA methods fall behind their predecessors when evaluated using these metrics, suggesting the necessity for a more balanced view of the performance of neural CO (NCO) solvers. We hope RL4CO will encourage the exploration of novel solutions to complex real-world tasks, allowing the NCO community to compare with existing methods through a standardized interface that decouples the science from software engineering. We make our library publicly available at //github.com/kaist-silab/rl4co.
Recent advancements in language models (LMs) have gained substantial attentions on their capability to generate human-like responses. Though exhibiting a promising future for various applications such as conversation AI, these LMs face deployment challenges on various devices due to their extreme computational cost and unpredictable inference latency. Such varied inference latency, identified as a consequence of uncertainty intrinsic to the nature of language, can lead to computational inefficiency and degrade the overall performance of LMs, especially under high-traffic workloads. Unfortunately, the bandwidth of these uncertainty sources is extensive, complicating the prediction of latency and the effects emanating from such uncertainties. To understand and mitigate the impact of uncertainty on real-time response-demanding systems, we take the first step to comprehend, quantify and optimize these uncertainty-induced latency performance variations in LMs. Specifically, we present RT-LM, an uncertainty-aware resource management ecosystem for real-time inference of LMs. RT-LM innovatively quantifies how specific input uncertainties, adversely affect latency, often leading to an increased output length. Exploiting these insights, we devise a lightweight yet effective method to dynamically correlate input text uncertainties with output length at runtime. Utilizing this quantification as a latency heuristic, we integrate the uncertainty information into a system-level scheduler which explores several uncertainty-induced optimization opportunities, including uncertainty-aware prioritization, dynamic consolidation, and strategic CPU offloading. Quantitative experiments across five state-of-the-art LMs on two hardware platforms demonstrates that RT-LM can significantly reduce the average response time and improve throughput while incurring a rather small runtime overhead.
Large language models (LLMs) have significantly advanced the field of natural language processing (NLP), providing a highly useful, task-agnostic foundation for a wide range of applications. The great promise of LLMs as general task solvers motivated people to extend their functionality largely beyond just a ``chatbot'', and use it as an assistant or even replacement for domain experts and tools in specific domains such as healthcare, finance, and education. However, directly applying LLMs to solve sophisticated problems in specific domains meets many hurdles, caused by the heterogeneity of domain data, the sophistication of domain knowledge, the uniqueness of domain objectives, and the diversity of the constraints (e.g., various social norms, cultural conformity, religious beliefs, and ethical standards in the domain applications). To fill such a gap, explosively-increase research, and practices have been conducted in very recent years on the domain specialization of LLMs, which, however, calls for a comprehensive and systematic review to better summarizes and guide this promising domain. In this survey paper, first, we propose a systematic taxonomy that categorizes the LLM domain-specialization techniques based on the accessibility to LLMs and summarizes the framework for all the subcategories as well as their relations and differences to each other. We also present a comprehensive taxonomy of critical application domains that can benefit from specialized LLMs, discussing their practical significance and open challenges. Furthermore, we offer insights into the current research status and future trends in this area.
The problem of answering questions using knowledge from pre-trained language models (LMs) and knowledge graphs (KGs) presents two challenges: given a QA context (question and answer choice), methods need to (i) identify relevant knowledge from large KGs, and (ii) perform joint reasoning over the QA context and KG. In this work, we propose a new model, QA-GNN, which addresses the above challenges through two key innovations: (i) relevance scoring, where we use LMs to estimate the importance of KG nodes relative to the given QA context, and (ii) joint reasoning, where we connect the QA context and KG to form a joint graph, and mutually update their representations through graph neural networks. We evaluate QA-GNN on the CommonsenseQA and OpenBookQA datasets, and show its improvement over existing LM and LM+KG models, as well as its capability to perform interpretable and structured reasoning, e.g., correctly handling negation in questions.
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