Parallel Decoding via Hidden Transfer for Lossless Large Language Model Acceleration

Large language models (LLMs) have recently shown remarkable performance across a wide range of tasks. However, the substantial number of parameters in LLMs contributes to significant latency during model inference. This is particularly evident when utilizing autoregressive decoding methods, which generate one token in a single forward process, thereby not fully capitalizing on the parallel computing capabilities of GPUs. In this paper, we propose a novel parallel decoding approach, namely \textit{hidden transfer}, which decodes multiple successive tokens simultaneously in a single forward pass. The idea is to transfer the intermediate hidden states of the previous context to the \textit{pseudo} hidden states of the future tokens to be generated, and then the pseudo hidden states will pass the following transformer layers thereby assimilating more semantic information and achieving superior predictive accuracy of the future tokens. Besides, we use the novel tree attention mechanism to simultaneously generate and verify multiple candidates of output sequences, which ensure the lossless generation and further improves the generation efficiency of our method. Experiments demonstrate the effectiveness of our method. We conduct a lot of analytic experiments to prove our motivation. In terms of acceleration metrics, we outperform all the single-model acceleration techniques, including Medusa and Self-Speculative decoding.

Q-PEFT: Query-dependent Parameter Efficient Fine-tuning for Text Reranking with Large Language Models

Parameter Efficient Fine-Tuning (PEFT) methods have been extensively utilized in Large Language Models (LLMs) to improve the down-streaming tasks without the cost of fine-tuing the whole LLMs. Recent studies have shown how to effectively use PEFT for fine-tuning LLMs in ranking tasks with convincing performance; there are some limitations, including the learned prompt being fixed for different documents, overfitting to specific tasks, and low adaptation ability. In this paper, we introduce a query-dependent parameter efficient fine-tuning (Q-PEFT) approach for text reranking to leak the information of the true queries to LLMs and then make the generation of true queries from input documents much easier. Specifically, we utilize the query to extract the top-$k$ tokens from concatenated documents, serving as contextual clues. We further augment Q-PEFT by substituting the retrieval mechanism with a multi-head attention layer to achieve end-to-end training and cover all the tokens in the documents, guiding the LLMs to generate more document-specific synthetic queries, thereby further improving the reranking performance. Extensive experiments are conducted on four public datasets, demonstrating the effectiveness of our proposed approach.

Think Twice Before Assure: Confidence Estimation for Large Language Models through Reflection on Multiple Answers

Confidence estimation aiming to evaluate output trustability is crucial for the application of large language models (LLM), especially the black-box ones. Existing confidence estimation of LLM is typically not calibrated due to the overconfidence of LLM on its generated incorrect answers. Existing approaches addressing the overconfidence issue are hindered by a significant limitation that they merely consider the confidence of one answer generated by LLM. To tackle this limitation, we propose a novel paradigm that thoroughly evaluates the trustability of multiple candidate answers to mitigate the overconfidence on incorrect answers. Building upon this paradigm, we introduce a two-step framework, which firstly instructs LLM to reflect and provide justifications for each answer, and then aggregates the justifications for comprehensive confidence estimation. This framework can be integrated with existing confidence estimation approaches for superior calibration. Experimental results on six datasets of three tasks demonstrate the rationality and effectiveness of the proposed framework.

Enhancing Long-Term Recommendation with Bi-level Learnable Large Language Model Planning

Traditional recommendation setting tends to excessively cater to users' immediate interests and neglect their long-term engagement. To address it, it is crucial to incorporate planning capabilities into the recommendation decision-making process to develop policies that take into account both immediate interests and long-term engagement. Despite Reinforcement Learning (RL) can learn planning capacity by maximizing cumulative reward, the scarcity of recommendation data presents challenges such as instability and susceptibility to overfitting when training RL models from scratch. In this context, we propose to leverage the remarkable planning capabilities over sparse data of Large Language Models (LLMs) for long-term recommendation. The key lies in enabling a language model to understand and apply task-solving principles effectively in personalized recommendation scenarios, as the model's pre-training may not naturally encompass these principles, necessitating the need to inspire or teach the model. To achieve this, we propose a Bi-level Learnable LLM Planner framework, which combines macro-learning and micro-learning through a hierarchical mechanism. The framework includes a Planner and Reflector for acquiring high-level guiding principles and an Actor-Critic component for planning personalization. Extensive experiments validate the superiority of the framework in learning to plan for long-term recommendations.

PDETime: Rethinking Long-Term Multivariate Time Series Forecasting from the perspective of partial differential equations

Recent advancements in deep learning have led to the development of various models for long-term multivariate time-series forecasting (LMTF), many of which have shown promising results. Generally, the focus has been on historical-value-based models, which rely on past observations to predict future series. Notably, a new trend has emerged with time-index-based models, offering a more nuanced understanding of the continuous dynamics underlying time series. Unlike these two types of models that aggregate the information of spatial domains or temporal domains, in this paper, we consider multivariate time series as spatiotemporal data regularly sampled from a continuous dynamical system, which can be represented by partial differential equations (PDEs), with the spatial domain being fixed. Building on this perspective, we present PDETime, a novel LMTF model inspired by the principles of Neural PDE solvers, following the encoding-integration-decoding operations. Our extensive experimentation across seven diverse real-world LMTF datasets reveals that PDETime not only adapts effectively to the intrinsic spatiotemporal nature of the data but also sets new benchmarks, achieving state-of-the-art results

Multimodal Instruction Tuning with Conditional Mixture of LoRA

Multimodal Large Language Models (MLLMs) have demonstrated remarkable proficiency in diverse tasks across different domains, with an increasing focus on improving their zero-shot generalization capabilities for unseen multimodal tasks. Multimodal instruction tuning has emerged as a successful strategy for achieving zero-shot generalization by fine-tuning pre-trained models on diverse multimodal tasks through instructions. As MLLMs grow in complexity and size, the need for parameter-efficient fine-tuning methods like Low-Rank Adaption (LoRA), which fine-tunes with a minimal set of parameters, becomes essential. However, applying LoRA in multimodal instruction tuning presents the challenge of task interference, which leads to performance degradation, especially when dealing with a broad array of multimodal tasks. To address this, this paper introduces a novel approach that integrates multimodal instruction tuning with Conditional Mixture-of-LoRA (MixLoRA). It innovates upon LoRA by dynamically constructing low-rank adaptation matrices tailored to the unique demands of each input instance, aiming to mitigate task interference. Experimental results on various multimodal evaluation datasets indicate that MixLoRA not only outperforms the conventional LoRA with the same or even higher ranks, demonstrating its efficacy and adaptability in diverse multimodal tasks.

Vision-Flan: Scaling Human-Labeled Tasks in Visual Instruction Tuning

Despite vision-language models' (VLMs) remarkable capabilities as versatile visual assistants, two substantial challenges persist within the existing VLM frameworks: (1) lacking task diversity in pretraining and visual instruction tuning, and (2) annotation error and bias in GPT-4 synthesized instruction tuning data. Both challenges lead to issues such as poor generalizability, hallucination, and catastrophic forgetting. To address these challenges, we construct Vision-Flan, the most diverse publicly available visual instruction tuning dataset to date, comprising 187 diverse tasks and 1,664,261 instances sourced from academic datasets, and each task is accompanied by an expert-written instruction. In addition, we propose a two-stage instruction tuning framework, in which VLMs are firstly finetuned on Vision-Flan and further tuned on GPT-4 synthesized data. We find this two-stage tuning framework significantly outperforms the traditional single-stage visual instruction tuning framework and achieves the state-of-the-art performance across a wide range of multi-modal evaluation benchmarks. Finally, we conduct in-depth analyses to understand visual instruction tuning and our findings reveal that: (1) GPT-4 synthesized data does not substantially enhance VLMs' capabilities but rather modulates the model's responses to human-preferred formats; (2) A minimal quantity (e.g., 1,000) of GPT-4 synthesized data can effectively align VLM responses with human-preference; (3) Visual instruction tuning mainly helps large-language models (LLMs) to understand visual features.

Navigating the Dual Facets: A Comprehensive Evaluation of Sequential Memory Editing in Large Language Models

Memory Editing (ME) has emerged as an efficient method to modify erroneous facts or inject new facts into Large Language Models (LLMs). Two mainstream ME methods exist: parameter-modifying ME and parameter-preserving ME (integrating extra modules while preserving original parameters). Regrettably, previous studies on ME evaluation have two critical limitations: (i) evaluating LLMs with single edit only, neglecting the need for continuous editing, and (ii) evaluations focusing solely on basic factual triples, overlooking broader LLM capabilities like logical reasoning and reading understanding. This study addresses these limitations with contributions threefold: (i) We explore how ME affects a wide range of fundamental capabilities of LLMs under sequential editing. Experimental results reveal an intriguing phenomenon: Most parameter-modifying ME consistently degrade performance across all tasks after a few sequential edits. In contrast, parameter-preserving ME effectively maintains LLMs' fundamental capabilities but struggles to accurately recall edited knowledge presented in a different format. (ii) We extend our evaluation to different editing settings, such as layers to edit, model size, instruction tuning, etc. Experimental findings indicate several strategies that can potentially mitigate the adverse effects of ME. (iii) We further explain why parameter-modifying ME damages LLMs from three dimensions: parameter changes after editing, language modeling capability, and the in-context learning capability. Our in-depth study advocates more careful use of ME in real-world scenarios.

Image Translation as Diffusion Visual Programmers

We introduce the novel Diffusion Visual Programmer (DVP), a neuro-symbolic image translation framework. Our proposed DVP seamlessly embeds a condition-flexible diffusion model within the GPT architecture, orchestrating a coherent sequence of visual programs (i.e., computer vision models) for various pro-symbolic steps, which span RoI identification, style transfer, and position manipulation, facilitating transparent and controllable image translation processes. Extensive experiments demonstrate DVP's remarkable performance, surpassing concurrent arts. This success can be attributed to several key features of DVP: First, DVP achieves condition-flexible translation via instance normalization, enabling the model to eliminate sensitivity caused by the manual guidance and optimally focus on textual descriptions for high-quality content generation. Second, the framework enhances in-context reasoning by deciphering intricate high-dimensional concepts in feature spaces into more accessible low-dimensional symbols (e.g., [Prompt], [RoI object]), allowing for localized, context-free editing while maintaining overall coherence. Last but not least, DVP improves systemic controllability and explainability by offering explicit symbolic representations at each programming stage, empowering users to intuitively interpret and modify results. Our research marks a substantial step towards harmonizing artificial image translation processes with cognitive intelligence, promising broader applications.