Dense Retrieval (DR) is now considered as a promising tool to enhance the memorization capacity of Large Language Models (LLM) such as GPT3 and GPT-4 by incorporating external memories. However, due to the paradigm discrepancy between text generation of LLM and DR, it is still an open challenge to integrate the retrieval and generation tasks in a shared LLM. In this paper, we propose an efficient LLM-Oriented Retrieval Tuner, namely LMORT, which decouples DR capacity from base LLM and non-invasively coordinates the optimally aligned and uniform layers of the LLM towards a unified DR space, achieving an efficient and effective DR without tuning the LLM itself. The extensive experiments on six BEIR datasets show that our approach could achieve competitive zero-shot retrieval performance compared to a range of strong DR models while maintaining the generation ability of LLM.
Large-scale pretrained language models have achieved compelling performance in a wide range of language understanding and information retrieval tasks. Knowledge distillation offers an opportunity to compress a large language model to a small one, in order to reach a reasonable latency-performance tradeoff. However, for scenarios where the number of requests (e.g., queries submitted to a search engine) is highly variant, the static tradeoff attained by the compressed language model might not always fit. Once a model is assigned with a static tradeoff, it could be inadequate in that the latency is too high when the number of requests is large or the performance is too low when the number of requests is small. To this end, we propose an elastic language model (ElasticLM) that elastically adjusts the tradeoff according to the request stream. The basic idea is to introduce a compute elasticity to the compressed language model, so that the tradeoff could vary on-the-fly along scalable and controllable compute. Specifically, we impose an elastic structure to enable ElasticLM with compute elasticity and design an elastic optimization to learn ElasticLM under compute elasticity. To serve ElasticLM, we apply an elastic schedule. Considering the specificity of information retrieval, we adapt ElasticLM to dense retrieval and reranking and present ElasticDenser and ElasticRanker respectively. Offline evaluation is conducted on a language understanding benchmark GLUE; and several information retrieval tasks including Natural Question, Trivia QA, and MS MARCO. The results show that ElasticLM along with ElasticDenser and ElasticRanker can perform correctly and competitively compared with an array of static baselines. Furthermore, online simulation with concurrency is also carried out. The results demonstrate that ElasticLM can provide elastic tradeoffs with respect to varying request stream.
Language model (LM) distillation is a trending area that aims to distil the knowledge resided in a large teacher LM to a small student one. While various methods have been proposed to push the distillation to its limits, it is still a pain distilling LMs when a large capacity gap is exhibited between the teacher and the student LMs. The pain is mainly resulted by the curse of capacity gap, which describes that a larger teacher LM cannot always lead to a better student LM than one distilled from a smaller teacher LM due to the affect of capacity gap increment. That is, there is likely an optimal point yielding the best student LM along the scaling course of the teacher LM. Even worse, the curse of capacity gap can be only partly yet not fully lifted as indicated in previous studies. However, the tale is not ever one-sided. Although a larger teacher LM has better performance than a smaller teacher LM, it is much more resource-demanding especially in the context of recent large LMs (LLMs). Consequently, instead of sticking to lifting the curse, leaving the curse as is should be arguably fine. Even better, in this paper, we reveal that the optimal capacity gap is almost consistent across different student scales and architectures, fortunately turning the curse into the law of capacity gap. The law later guides us to distil a 3B student LM (termed MiniMA) from a 7B teacher LM (adapted LLaMA2-7B). MiniMA is demonstrated to yield a new compute-performance pareto frontier among existing 3B LMs on commonly used benchmarks, and its instruction-tuned version (termed MiniChat) outperforms a wide range of 3B competitors in GPT4 evaluation and could even compete with several 7B chat models.
Recently, SimCSE has shown the feasibility of contrastive learning in training sentence embeddings and illustrates its expressiveness in spanning an aligned and uniform embedding space. However, prior studies have shown that dense models could contain harmful parameters that affect the model performance, and it is no wonder that SimCSE can as well be invented with such parameters. Driven by this, parameter sparsification is applied, where alignment and uniformity scores are used to measure the contribution of each parameter to the overall quality of sentence embeddings. Drawing from a preliminary study, we consider parameters with minimal contributions to be detrimental, as their sparsification results in improved model performance. To discuss the ubiquity of detrimental parameters and remove them, more experiments on the standard semantic textual similarity (STS) tasks and transfer learning tasks are conducted, and the results show that the proposed sparsified SimCSE (SparseCSE) has excellent performance in comparison with SimCSE. Furthermore, through in-depth analysis, we establish the validity and stability of our sparsification method, showcasing that the embedding space generated by SparseCSE exhibits improved alignment compared to that produced by SimCSE. Importantly, the uniformity yet remains uncompromised.
Large-scale Causal Language Models (CLMs), e.g., GPT3 and ChatGPT, have brought great success in text generation. However, it is still an open challenge to control the generation process of CLM while balancing flexibility, control granularity, and generation efficiency. In this paper, we provide a new alternative for controllable text generation (CTG), by designing a non-intrusive, lightweight control plugin to accompany the generation of CLM at arbitrary time steps. The proposed control plugin, namely Residual Memory Transformer (RMT), has an encoder-decoder setup, which can accept any types of control conditions and cooperate with CLM through a residual learning paradigm, to achieve a more flexible, general, and efficient CTG. Extensive experiments are carried out on various control tasks, in the form of both automatic and human evaluations. The results show the superiority of RMT over a range of state-of-the-art approaches, proving the effectiveness and versatility of our approach.
Sarcasm, sentiment, and emotion are three typical kinds of spontaneous affective responses of humans to external events and they are tightly intertwined with each other. Such events may be expressed in multiple modalities (e.g., linguistic, visual and acoustic), e.g., multi-modal conversations. Joint analysis of humans' multi-modal sarcasm, sentiment, and emotion is an important yet challenging topic, as it is a complex cognitive process involving both cross-modality interaction and cross-affection correlation. From the probability theory perspective, cross-affection correlation also means that the judgments on sarcasm, sentiment, and emotion are incompatible. However, this exposed phenomenon cannot be sufficiently modelled by classical probability theory due to its assumption of compatibility. Neither do the existing approaches take it into consideration. In view of the recent success of quantum probability (QP) in modeling human cognition, particularly contextual incompatible decision making, we take the first step towards introducing QP into joint multi-modal sarcasm, sentiment, and emotion analysis. Specifically, we propose a QUantum probabIlity driven multi-modal sarcasm, sEntiment and emoTion analysis framework, termed QUIET. Extensive experiments on two datasets and the results show that the effectiveness and advantages of QUIET in comparison with a wide range of the state-of-the-art baselines. We also show the great potential of QP in multi-affect analysis.
Quantum theory, originally proposed as a physical theory to describe the motions of microscopic particles, has been applied to various non-physics domains involving human cognition and decision-making that are inherently uncertain and exhibit certain non-classical, quantum-like characteristics. Sentiment analysis is a typical example of such domains. In the last few years, by leveraging the modeling power of quantum probability (a non-classical probability stemming from quantum mechanics methodology) and deep neural networks, a range of novel quantum-cognitively inspired models for sentiment analysis have emerged and performed well. This survey presents a timely overview of the latest developments in this fascinating cross-disciplinary area. We first provide a background of quantum probability and quantum cognition at a theoretical level, analyzing their advantages over classical theories in modeling the cognitive aspects of sentiment analysis. Then, recent quantum-cognitively inspired models are introduced and discussed in detail, focusing on how they approach the key challenges of the sentiment analysis task. Finally, we discuss the limitations of the current research and highlight future research directions.
Finetuning pretrained language models (LMs) have enabled appealing performance on a diverse array of tasks. The intriguing task-agnostic property has driven a shifted focus from task-specific to task-agnostic distillation of LMs. While task-agnostic, compute-efficient, performance-preserved LMs can be yielded by task-agnostic distillation, previous studies mainly sit in distillation of either encoder-only LMs (e.g., BERT) or decoder-only ones (e.g., GPT) yet largely neglect that distillation of encoder-decoder LMs (e.g., T5) can posit very distinguished behaviors. Frustratingly, we discover that existing task-agnostic distillation methods can fail to handle the distillation of encoder-decoder LMs. To the demand, we explore a few paths and uncover a path named as MiniEnD that successfully tackles the distillation of encoder-decoder LMs in a task-agnostic fashion. We examine MiniEnD on language understanding and abstractive summarization. The results showcase that MiniEnD is generally effective and is competitive compared to other alternatives. We further scale MiniEnD up to distillation of 3B encoder-decoder language models with interpolated distillation. The results imply the opportunities and challenges in distilling large language models (e.g., LLaMA).
Pretrained language models (LMs) have shown compelling performance on various downstream tasks, but unfortunately they require a tremendous amount of inference compute. Knowledge distillation finds a path to compress LMs to small ones with a teacher-student paradigm. However, when the capacity gap between the teacher and the student is large, a curse of capacity gap appears, invoking a deficiency in distilling LMs. While a few studies have been carried out to fill the gap, the curse is not yet well tackled. In this paper, we aim at lifting the curse of capacity gap via enlarging the capacity of the student without notably increasing the inference compute. Largely motivated by sparse activation regime of mixture of experts (MoE), we propose a mixture of minimal experts (MiniMoE), which imposes extra parameters to the student but introduces almost no additional inference compute. Experimental results on GLUE and CoNLL demonstrate the curse of capacity gap is lifted by the magic of MiniMoE to a large extent. MiniMoE also achieves the state-of-the-art performance at small FLOPs compared with a range of competitive baselines. With a compression rate as much as $\sim$50$\times$, MiniMoE preserves $\sim$95\% GLUE score of the teacher.
New retrieval tasks have always been emerging, thus urging the development of new retrieval models. However, instantiating a retrieval model for each new retrieval task is resource-intensive and time-consuming, especially for a retrieval model that employs a large-scale pre-trained language model. To address this issue, we shift to a novel retrieval paradigm called modular retrieval, which aims to solve new retrieval tasks by instead composing multiple existing retrieval modules. Built upon the paradigm, we propose a retrieval model with modular prompt tuning named REMOP. It constructs retrieval modules subject to task attributes with deep prompt tuning, and yields retrieval models subject to tasks with module composition. We validate that, REMOP inherently with modularity not only has appealing generalizability and interpretability in preliminary explorations, but also achieves comparable performance to state-of-the-art retrieval models on a zero-shot retrieval benchmark.\footnote{Our code is available at \url{https://github.com/FreedomIntelligence/REMOP}}