Breaking Memory Limits: Gradient Wavelet Transform Enhances LLMs Training

Large language models (LLMs) have shown impressive performance across a range of natural language processing tasks. However, their vast number of parameters introduces significant memory challenges during training, particularly when using memory-intensive optimizers like Adam. Existing memory-efficient algorithms often rely on techniques such as singular value decomposition projection or weight freezing. While these approaches help alleviate memory constraints, they generally produce suboptimal results compared to full-rank updates. In this paper, we investigate the memory-efficient method beyond low-rank training, proposing a novel solution called Gradient Wavelet Transform (GWT), which applies wavelet transforms to gradients in order to significantly reduce the memory requirements for maintaining optimizer states. We demonstrate that GWT can be seamlessly integrated with memory-intensive optimizers, enabling efficient training without sacrificing performance. Through extensive experiments on both pre-training and fine-tuning tasks, we show that GWT achieves state-of-the-art performance compared with advanced memory-efficient optimizers and full-rank approaches in terms of both memory usage and training performance.

Dovetail: A CPU/GPU Heterogeneous Speculative Decoding for LLM inference

Due to the high resource demands of Large Language Models (LLMs), achieving widespread deployment on consumer-grade devices presents significant challenges. Typically, personal or consumer-grade devices, including servers configured prior to the era of large-scale models, generally have relatively weak GPUs and relatively strong CPUs. However, most current methods primarily depend on GPUs for computation. Therefore, we propose Dovetail, an approach that deploys the draft model on the GPU to generate draft tokens while allowing the target model to perform parallel verification on the CPU, thereby improving the utilization of all available hardware resources and occupying less inter-device communication bandwidth. Accordingly, we have redesigned the draft model to better align with heterogeneous hardware characteristics. To this end, we implemented several optimizations: reducing the number of draft tokens to mitigate latency in parallel verification, increasing the depth of the draft model to enhance its predictive capacity, and introducing DGF (Dynamic Gating Fusion) to improve the integration of features and token embeddings. In the HumanEval benchmark, Dovetail achieved an inference speed of 5.86 tokens per second for LLaMA2-Chat-7B using 3GB of VRAM, representing an approximately 2.77x improvement over CPU-only inference. Furthermore, the inference speed was increased to 8 tokens per second when utilizing 7GB of VRAM.

SCBench: A KV Cache-Centric Analysis of Long-Context Methods

Long-context LLMs have enabled numerous downstream applications but also introduced significant challenges related to computational and memory efficiency. To address these challenges, optimizations for long-context inference have been developed, centered around the KV cache. However, existing benchmarks often evaluate in single-request, neglecting the full lifecycle of the KV cache in real-world use. This oversight is particularly critical, as KV cache reuse has become widely adopted in LLMs inference frameworks, such as vLLM and SGLang, as well as by LLM providers, including OpenAI, Microsoft, Google, and Anthropic. To address this gap, we introduce SCBench(SharedContextBench), a comprehensive benchmark for evaluating long-context methods from a KV cachecentric perspective: 1) KV cache generation, 2) KV cache compression, 3) KV cache retrieval, 4) KV cache loading. Specifically, SCBench uses test examples with shared context, ranging 12 tasks with two shared context modes, covering four categories of long-context capabilities: string retrieval, semantic retrieval, global information, and multi-task. With it, we provide an extensive KV cache-centric analysis of eight categories long-context solutions, including Gated Linear RNNs, Mamba-Attention hybrids, and efficient methods such as sparse attention, KV cache dropping, quantization, retrieval, loading, and prompt compression. The evaluation is conducted on 8 long-context LLMs. Our findings show that sub-O(n) memory methods suffer in multi-turn scenarios, while sparse encoding with O(n) memory and sub-O(n^2) pre-filling computation perform robustly. Dynamic sparsity yields more expressive KV caches than static patterns, and layer-level sparsity in hybrid architectures reduces memory usage with strong performance. Additionally, we identify attention distribution shift issues in long-generation scenarios. https://aka.ms/SCBench.

Oracle-guided Dynamic User Preference Modeling for Sequential Recommendation

Sequential recommendation methods can capture dynamic user preferences from user historical interactions to achieve better performance. However, most existing methods only use past information extracted from user historical interactions to train the models, leading to the deviations of user preference modeling. Besides past information, future information is also available during training, which contains the ``oracle'' user preferences in the future and will be beneficial to model dynamic user preferences. Therefore, we propose an oracle-guided dynamic user preference modeling method for sequential recommendation (Oracle4Rec), which leverages future information to guide model training on past information, aiming to learn ``forward-looking'' models. Specifically, Oracle4Rec first extracts past and future information through two separate encoders, then learns a forward-looking model through an oracle-guiding module which minimizes the discrepancy between past and future information. We also tailor a two-phase model training strategy to make the guiding more effective. Extensive experiments demonstrate that Oracle4Rec is superior to state-of-the-art sequential methods. Further experiments show that Oracle4Rec can be leveraged as a generic module in other sequential recommendation methods to improve their performance with a considerable margin.

Acceleration for Deep Reinforcement Learning using Parallel and Distributed Computing: A Survey

Deep reinforcement learning has led to dramatic breakthroughs in the field of artificial intelligence for the past few years. As the amount of rollout experience data and the size of neural networks for deep reinforcement learning have grown continuously, handling the training process and reducing the time consumption using parallel and distributed computing is becoming an urgent and essential desire. In this paper, we perform a broad and thorough investigation on training acceleration methodologies for deep reinforcement learning based on parallel and distributed computing, providing a comprehensive survey in this field with state-of-the-art methods and pointers to core references. In particular, a taxonomy of literature is provided, along with a discussion of emerging topics and open issues. This incorporates learning system architectures, simulation parallelism, computing parallelism, distributed synchronization mechanisms, and deep evolutionary reinforcement learning. Further, we compare 16 current open-source libraries and platforms with criteria of facilitating rapid development. Finally, we extrapolate future directions that deserve further research.

Constructing and Masking Preference Profile with LLMs for Filtering Discomforting Recommendation

Personalized algorithms can inadvertently expose users to discomforting recommendations, potentially triggering negative consequences. The subjectivity of discomfort and the black-box nature of these algorithms make it challenging to effectively identify and filter such content. To address this, we first conducted a formative study to understand users' practices and expectations regarding discomforting recommendation filtering. Then, we designed a Large Language Model (LLM)-based tool named DiscomfortFilter, which constructs an editable preference profile for a user and helps the user express filtering needs through conversation to mask discomforting preferences within the profile. Based on the edited profile, DiscomfortFilter facilitates the discomforting recommendations filtering in a plug-and-play manner, maintaining flexibility and transparency. The constructed preference profile improves LLM reasoning and simplifies user alignment, enabling a 3.8B open-source LLM to rival top commercial models in an offline proxy task. A one-week user study with 24 participants demonstrated the effectiveness of DiscomfortFilter, while also highlighting its potential impact on platform recommendation outcomes. We conclude by discussing the ongoing challenges, highlighting its relevance to broader research, assessing stakeholder impact, and outlining future research directions.

How Do Large Language Models Understand Graph Patterns? A Benchmark for Graph Pattern Comprehension

Benchmarking the capabilities and limitations of large language models (LLMs) in graph-related tasks is becoming an increasingly popular and crucial area of research. Recent studies have shown that LLMs exhibit a preliminary ability to understand graph structures and node features. However, the potential of LLMs in graph pattern mining remains largely unexplored. This is a key component in fields such as computational chemistry, biology, and social network analysis. To bridge this gap, this work introduces a comprehensive benchmark to assess LLMs' capabilities in graph pattern tasks. We have developed a benchmark that evaluates whether LLMs can understand graph patterns based on either terminological or topological descriptions. Additionally, our benchmark tests the LLMs' capacity to autonomously discover graph patterns from data. The benchmark encompasses both synthetic and real datasets, and a variety of models, with a total of 11 tasks and 7 models. Our experimental framework is designed for easy expansion to accommodate new models and datasets. Our findings reveal that: (1) LLMs have preliminary abilities to understand graph patterns, with O1-mini outperforming in the majority of tasks; (2) Formatting input data to align with the knowledge acquired during pretraining can enhance performance; (3) The strategies employed by LLMs may differ from those used in conventional algorithms.

Translating Mental Imaginations into Characters with Codebooks and Dynamics-Enhanced Decoding

Advancements in non-invasive electroencephalogram (EEG)-based Brain-Computer Interface (BCI) technology have enabled communication through brain activity, offering significant potential for individuals with motor impairments. Existing methods for decoding characters or words from EEG recordings either rely on continuous external stimulation for high decoding accuracy or depend on direct intention imagination, which suffers from reduced discrimination ability. To overcome these limitations, we introduce a novel EEG paradigm based on mental tasks that achieves high discrimination accuracy without external stimulation. Specifically, we propose a codebook in which each letter or number is associated with a unique code that integrates three mental tasks, interleaved with eye-open and eye-closed states. This approach allows individuals to internally reference characters without external stimuli while maintaining reasonable accuracy. For enhanced decoding performance, we apply a Temporal-Spatial-Latent-Dynamics (TSLD) network to capture latent dynamics of spatiotemporal EEG signals. Experimental results demonstrate the effectiveness of our proposed EEG paradigm which achieves five times higher accuracy over direct imagination. Additionally, the TSLD network improves baseline methods by approximately 8.5%. Further more, we observe consistent performance improvement throughout the data collection process, suggesting that the proposed paradigm has potential for further optimization with continued use.

Zero-resource Hallucination Detection for Text Generation via Graph-based Contextual Knowledge Triples Modeling

LLMs obtain remarkable performance but suffer from hallucinations. Most research on detecting hallucination focuses on the questions with short and concrete correct answers that are easy to check the faithfulness. Hallucination detections for text generation with open-ended answers are more challenging. Some researchers use external knowledge to detect hallucinations in generated texts, but external resources for specific scenarios are hard to access. Recent studies on detecting hallucinations in long text without external resources conduct consistency comparison among multiple sampled outputs. To handle long texts, researchers split long texts into multiple facts and individually compare the consistency of each pairs of facts. However, these methods (1) hardly achieve alignment among multiple facts; (2) overlook dependencies between multiple contextual facts. In this paper, we propose a graph-based context-aware (GCA) hallucination detection for text generations, which aligns knowledge facts and considers the dependencies between contextual knowledge triples in consistency comparison. Particularly, to align multiple facts, we conduct a triple-oriented response segmentation to extract multiple knowledge triples. To model dependencies among contextual knowledge triple (facts), we construct contextual triple into a graph and enhance triples' interactions via message passing and aggregating via RGCN. To avoid the omission of knowledge triples in long text, we conduct a LLM-based reverse verification via reconstructing the knowledge triples. Experiments show that our model enhances hallucination detection and excels all baselines.

Federated Prediction-Powered Inference from Decentralized Data

In various domains, the increasing application of machine learning allows researchers to access inexpensive predictive data, which can be utilized as auxiliary data for statistical inference. Although such data are often unreliable compared to gold-standard datasets, Prediction-Powered Inference (PPI) has been proposed to ensure statistical validity despite the unreliability. However, the challenge of `data silos' arises when the private gold-standard datasets are non-shareable for model training, leading to less accurate predictive models and invalid inferences. In this paper, we introduces the Federated Prediction-Powered Inference (Fed-PPI) framework, which addresses this challenge by enabling decentralized experimental data to contribute to statistically valid conclusions without sharing private information. The Fed-PPI framework involves training local models on private data, aggregating them through Federated Learning (FL), and deriving confidence intervals using PPI computation. The proposed framework is evaluated through experiments, demonstrating its effectiveness in producing valid confidence intervals.