Abstract:Transformer-based foundation models (FMs) have recently demonstrated remarkable performance in medical image segmentation. However, scaling these models is challenging due to the limited size of medical image datasets within isolated hospitals, where data centralization is restricted due to privacy concerns. These constraints, combined with the data-intensive nature of FMs, hinder their broader application. Integrating federated learning (FL) with foundation models (FLFM) fine-tuning offers a potential solution to these challenges by enabling collaborative model training without data sharing, thus allowing FMs to take advantage of a diverse pool of sensitive medical image data across hospitals/clients. However, non-independent and identically distributed (non-IID) data among clients, paired with computational and communication constraints in federated environments, presents an additional challenge that limits further performance improvements and remains inadequately addressed in existing studies. In this work, we propose a novel FLFM fine-tuning framework, \underline{\textbf{Fed}}erated tuning with \underline{\textbf{S}}imilarity-guided \underline{\textbf{C}}ollaborative \underline{\textbf{A}}ggregation (FedSCA), encompassing all phases of the FL process. This includes (1) specially designed parameter-efficient fine-tuning (PEFT) for local client training to enhance computational efficiency; (2) partial low-level adapter transmission for communication efficiency; and (3) similarity-guided collaborative aggregation (SGCA) on the server side to address non-IID issues. Extensive experiments on three FL benchmarks for medical image segmentation demonstrate the effectiveness of our proposed FedSCA, establishing new SOTA performance.
Abstract:User modeling in large e-commerce platforms aims to optimize user experiences by incorporating various customer activities. Traditional models targeting a single task often focus on specific business metrics, neglecting the comprehensive user behavior, and thus limiting their effectiveness. To develop more generalized user representations, some existing work adopts Multi-task Learning (MTL)approaches. But they all face the challenges of optimization imbalance and inefficiency in adapting to new tasks. Continual Learning (CL), which allows models to learn new tasks incrementally and independently, has emerged as a solution to MTL's limitations. However, CL faces the challenge of catastrophic forgetting, where previously learned knowledge is lost when the model is learning the new task. Inspired by the success of prompt tuning in Pretrained Language Models (PLMs), we propose PCL, a Prompt-based Continual Learning framework for user modeling, which utilizes position-wise prompts as external memory for each task, preserving knowledge and mitigating catastrophic forgetting. Additionally, we design contextual prompts to capture and leverage inter-task relationships during prompt tuning. We conduct extensive experiments on real-world datasets to demonstrate PCL's effectiveness.
Abstract:Metaphor detection, a critical task in natural language processing, involves identifying whether a particular word in a sentence is used metaphorically. Traditional approaches often rely on supervised learning models that implicitly encode semantic relationships based on metaphor theories. However, these methods often suffer from a lack of transparency in their decision-making processes, which undermines the reliability of their predictions. Recent research indicates that LLMs (large language models) exhibit significant potential in metaphor detection. Nevertheless, their reasoning capabilities are constrained by predefined knowledge graphs. To overcome these limitations, we propose DMD, a novel dual-perspective framework that harnesses both implicit and explicit applications of metaphor theories to guide LLMs in metaphor detection and adopts a self-judgment mechanism to validate the responses from the aforementioned forms of guidance. In comparison to previous methods, our framework offers more transparent reasoning processes and delivers more reliable predictions. Experimental results prove the effectiveness of DMD, demonstrating state-of-the-art performance across widely-used datasets.
Abstract:Large Language models (LLMs) have become a research hotspot. To accelerate the inference of LLMs, storing computed caches in memory has become the standard technique. However, as the inference length increases, growing KV caches might lead to out-of-memory issues. Many existing methods address this issue through KV cache compression, primarily by preserving key tokens throughout all layers to reduce information loss. Most of them allocate a uniform budget size for each layer to retain. However, we observe that the minimum budget sizes needed to retain essential information vary across layers and models based on the perspectives of attention and hidden state output. Building on this observation, this paper proposes a simple yet effective KV cache compression method that leverages layer uncertainty to allocate budget size for each layer. Experimental results show that the proposed method can reduce memory usage of the KV caches to only $\sim$20\% when compared to Full KV inference while achieving nearly lossless performance.
Abstract:Dense retrieval in most industries employs dual-tower architectures to retrieve query-relevant documents. Due to online deployment requirements, existing real-world dense retrieval systems mainly enhance performance by designing negative sampling strategies, overlooking the advantages of scaling up. Recently, Large Language Models (LLMs) have exhibited superior performance that can be leveraged for scaling up dense retrieval. However, scaling up retrieval models significantly increases online query latency. To address this challenge, we propose ScalingNote, a two-stage method to exploit the scaling potential of LLMs for retrieval while maintaining online query latency. The first stage is training dual towers, both initialized from the same LLM, to unlock the potential of LLMs for dense retrieval. Then, we distill only the query tower using mean squared error loss and cosine similarity to reduce online costs. Through theoretical analysis and comprehensive offline and online experiments, we show the effectiveness and efficiency of ScalingNote. Our two-stage scaling method outperforms end-to-end models and verifies the scaling law of dense retrieval with LLMs in industrial scenarios, enabling cost-effective scaling of dense retrieval systems. Our online method incorporating ScalingNote significantly enhances the relevance between retrieved documents and queries.
Abstract:Scaling large language models (LLMs) demands extensive data and computing resources, which are traditionally constrained to data centers by the high-bandwidth requirements of distributed training. Low-bandwidth methods like federated learning (FL) could enable collaborative training of larger models across weakly-connected GPUs if they can effectively be used for pre-training. To achieve this, we introduce Photon, the first complete system for federated end-to-end LLM training, leveraging cross-silo FL for global-scale training with minimal communication overheads. Using Photon, we train the first federated family of decoder-only LLMs from scratch. We show that: (1) Photon can train model sizes up to 7B in a federated fashion while reaching an even better perplexity than centralized pre-training; (2) Photon model training time decreases with available compute, achieving a similar compute-time trade-off to centralized; and (3) Photon outperforms the wall-time of baseline distributed training methods by 35% via communicating 64x-512xless. Our proposal is robust to data heterogeneity and converges twice as fast as previous methods like DiLoCo. This surprising data efficiency stems from a unique approach combining small client batch sizes with extremely high learning rates, enabled by federated averaging's robustness to hyperparameters. Photon thus represents the first economical system for global internet-wide LLM pre-training.
Abstract:Long-context efficiency has recently become a trending topic in serving large language models (LLMs). And mixture of depths (MoD) is proposed as a perfect fit to bring down both latency and memory. In this paper, however, we discover that MoD can barely transform existing LLMs without costly training over an extensive number of tokens. To enable the transformations from any LLMs to MoD ones, we showcase top-k operator in MoD should be promoted to threshold-p operator, and refinement to architecture and data should also be crafted along. All these designs form our method termed MoDification. Through a comprehensive set of experiments covering model scales from 3B to 70B, we exhibit MoDification strikes an excellent balance between efficiency and effectiveness. MoDification can achieve up to ~1.2x speedup in latency and ~1.8x reduction in memory compared to original LLMs especially in long-context applications.
Abstract:Prompt engineering is very important to enhance the performance of large language models (LLMs). When dealing with complex issues, prompt engineers tend to distill multiple patterns from examples and inject relevant solutions to optimize the prompts, achieving satisfying results. However, existing automatic prompt optimization techniques are only limited to producing single flow instructions, struggling with handling diverse patterns. In this paper, we present AMPO, an automatic prompt optimization method that can iteratively develop a multi-branched prompt using failure cases as feedback. Our goal is to explore a novel way of structuring prompts with multi-branches to better handle multiple patterns in complex tasks, for which we introduce three modules: Pattern Recognition, Branch Adjustment, and Branch Pruning. In experiments across five tasks, AMPO consistently achieves the best results. Additionally, our approach demonstrates significant optimization efficiency due to our adoption of a minimal search strategy.
Abstract:Prompt engineering is pivotal for harnessing the capabilities of large language models (LLMs) across diverse applications. While existing prompt optimization methods improve prompt effectiveness, they often lead to prompt drifting, where newly generated prompts can adversely impact previously successful cases while addressing failures. Furthermore, these methods tend to rely heavily on LLMs' intrinsic capabilities for prompt optimization tasks. In this paper, we introduce StraGo (Strategic-Guided Optimization), a novel approach designed to mitigate prompt drifting by leveraging insights from both successful and failed cases to identify critical factors for achieving optimization objectives. StraGo employs a how-to-do methodology, integrating in-context learning to formulate specific, actionable strategies that provide detailed, step-by-step guidance for prompt optimization. Extensive experiments conducted across a range of tasks, including reasoning, natural language understanding, domain-specific knowledge, and industrial applications, demonstrate StraGo's superior performance. It establishes a new state-of-the-art in prompt optimization, showcasing its ability to deliver stable and effective prompt improvements.
Abstract:Language Model pre-training benefits from a broader data mixture to enhance performance across domains and languages. However, training on such heterogeneous text corpora is complex, requiring extensive and cost-intensive efforts. Since these data sources vary in lexical, syntactic, and semantic aspects, they cause negative interference or the "curse of multilinguality". We propose a novel pre-training framework to alleviate this curse. Our method, DEPT, decouples the embedding layers from the transformer body while simultaneously training the latter in multiple contexts. DEPT enables the model to train without being bound to a shared global vocabulary. DEPT: (1) can train robustly and effectively under significant data heterogeneity, (2) reduces the parameter count of the token embeddings by up to 80% and the communication costs by 675x for billion-scale models (3) enhances model generalization and plasticity in adapting to new languages and domains, and (4) allows training with custom optimized vocabulary per data source. We prove DEPT's potential by performing the first vocabulary-agnostic federated multilingual pre-training of a 1.3 billion-parameter model across high and low-resource languages, reducing its parameter count by 409 million.