Zhenyu Zhang

Harry Dong, Xinyu Yang, Zhenyu Zhang et al.

Many computational factors limit broader deployment of large language models. In this paper, we focus on a memory bottleneck imposed by the key-value (KV) cache, a computational shortcut that requires storing previous KV pairs during decoding. While existing KV cache methods approach this problem by pruning or evicting large swaths of relatively less important KV pairs to dramatically reduce the memory footprint of the cache, they can have limited success in tasks that require recollecting a majority of previous tokens. To alleviate this issue, we propose LESS, a simple integration of a (nearly free) constant sized cache with eviction-based cache methods, such that all tokens can be queried at later decoding steps. Its ability to retain information throughout time shows merit on a variety of tasks where we demonstrate LESS can help reduce the performance gap from caching everything, sometimes even matching it, all while being efficient. Relevant code can be found at https://github.com/hdong920/LESS.

Jiawei Zhao, Zhenyu Zhang, Beidi Chen et al.

Training Large Language Models (LLMs) presents significant memory challenges, predominantly due to the growing size of weights and optimizer states. Common memory-reduction approaches, such as low-rank adaptation (LoRA), add a trainable low-rank matrix to the frozen pre-trained weight in each layer, reducing trainable parameters and optimizer states. However, such approaches typically underperform training with full-rank weights in both pre-training and fine-tuning stages since they limit the parameter search to a low-rank subspace and alter the training dynamics, and further, may require full-rank warm start. In this work, we propose Gradient Low-Rank Projection (GaLore), a training strategy that allows full-parameter learning but is more memory-efficient than common low-rank adaptation methods such as LoRA. Our approach reduces memory usage by up to 65.5% in optimizer states while maintaining both efficiency and performance for pre-training on LLaMA 1B and 7B architectures with C4 dataset with up to 19.7B tokens, and on fine-tuning RoBERTa on GLUE tasks. Our 8-bit GaLore further reduces optimizer memory by up to 82.5% and total training memory by 63.3%, compared to a BF16 baseline. Notably, we demonstrate, for the first time, the feasibility of pre-training a 7B model on consumer GPUs with 24GB memory (e.g., NVIDIA RTX 4090) without model parallel, checkpointing, or offloading strategies.

Yeonju Ro, Zhenyu Zhang, Souvik Kundu et al.

Large language models (LLMs) excel at capturing global token dependencies via self-attention but face prohibitive compute and memory costs on lengthy inputs. While sub-quadratic methods (e.g., linear attention) can reduce these costs, they often degrade accuracy due to overemphasizing recent tokens. In this work, we first propose dual-state linear attention (DSLA), a novel design that maintains two specialized hidden states-one for preserving historical context and one for tracking recency-thereby mitigating the short-range bias typical of linear-attention architectures. To further balance efficiency and accuracy under dynamic workload conditions, we introduce DSLA-Serve, an online adaptive distillation framework that progressively replaces Transformer layers with DSLA layers at inference time, guided by a sensitivity-based layer ordering. DSLA-Serve uses a chained fine-tuning strategy to ensure that each newly converted DSLA layer remains consistent with previously replaced layers, preserving the overall quality. Extensive evaluations on commonsense reasoning, long-context QA, and text summarization demonstrate that DSLA-Serve yields 2.3x faster inference than Llama2-7B and 3.0x faster than the hybrid Zamba-7B, while retaining comparable performance across downstream tasks. Our ablation studies show that DSLA's dual states capture both global and local dependencies, addressing the historical-token underrepresentation seen in prior linear attentions. Codes are available at https://github.com/utnslab/DSLA-Serve.

Yilong Chen, Junyuan Shang, Zhenyu Zhang et al.

Large language models (LLMs) face inherent performance bottlenecks under parameter constraints, particularly in processing critical tokens that demand complex reasoning. Empirical analysis reveals challenging tokens induce abrupt gradient spikes across layers, exposing architectural stress points in standard Transformers. Building on this insight, we propose Inner Thinking Transformer (ITT), which reimagines layer computations as implicit thinking steps. ITT dynamically allocates computation through Adaptive Token Routing, iteratively refines representations via Residual Thinking Connections, and distinguishes reasoning phases using Thinking Step Encoding. ITT enables deeper processing of critical tokens without parameter expansion. Evaluations across 162M-466M parameter models show ITT achieves 96.5\% performance of a 466M Transformer using only 162M parameters, reduces training data by 43.2\%, and outperforms Transformer/Loop variants in 11 benchmarks. By enabling elastic computation allocation during inference, ITT balances performance and efficiency through architecture-aware optimization of implicit thinking pathways.

Naibin Gu, Yilong Chen, Zhenyu Zhang et al.

Although scaling up the number of trainable parameters in both pre-training and fine-tuning can effectively improve the performance of large language models, it also leads to increased computational overhead. When delving into the parameter difference, we find that a subset of parameters, termed advantageous parameters, plays a crucial role in determining model performance. Further analysis reveals that stronger models tend to possess more such parameters. In this paper, we propose Advantageous Parameter EXpansion Training (APEX), a method that progressively expands advantageous parameters into the space of disadvantageous ones, thereby increasing their proportion and enhancing training effectiveness. Further theoretical analysis from the perspective of matrix effective rank explains the performance gains of APEX. Extensive experiments on both instruction tuning and continued pre-training demonstrate that, in instruction tuning, APEX outperforms full-parameter tuning while using only 52% of the trainable parameters. In continued pre-training, APEX achieves the same perplexity level as conventional training with just 33% of the training data, and yields significant improvements on downstream tasks.

Guangba Yu, Gou Tan, Haojia Huang et al.

The rapid advancement of Artificial Intelligence (AI) has led to its integration into various areas, especially with Large Language Models (LLMs) significantly enhancing capabilities in Artificial Intelligence Generated Content (AIGC). However, the complexity of AI systems has also exposed their vulnerabilities, necessitating robust methods for failure analysis (FA) and fault injection (FI) to ensure resilience and reliability. Despite the importance of these techniques, there lacks a comprehensive review of FA and FI methodologies in AI systems. This study fills this gap by presenting a detailed survey of existing FA and FI approaches across six layers of AI systems. We systematically analyze 160 papers and repositories to answer three research questions including (1) what are the prevalent failures in AI systems, (2) what types of faults can current FI tools simulate, (3) what gaps exist between the simulated faults and real-world failures. Our findings reveal a taxonomy of AI system failures, assess the capabilities of existing FI tools, and highlight discrepancies between real-world and simulated failures. Moreover, this survey contributes to the field by providing a framework for fault diagnosis, evaluating the state-of-the-art in FI, and identifying areas for improvement in FI techniques to enhance the resilience of AI systems.

Yilong Chen, Linhao Zhang, Junyuan Shang et al.

Large language models (LLMs) with billions of parameters demonstrate impressive performance. However, the widely used Multi-Head Attention (MHA) in LLMs incurs substantial computational and memory costs during inference. While some efforts have optimized attention mechanisms by pruning heads or sharing parameters among heads, these methods often lead to performance degradation or necessitate substantial continued pre-training costs to restore performance. Based on the analysis of attention redundancy, we design a Decoupled-Head Attention (DHA) mechanism. DHA adaptively configures group sharing for key heads and value heads across various layers, achieving a better balance between performance and efficiency. Inspired by the observation of clustering similar heads, we propose to progressively transform the MHA checkpoint into the DHA model through linear fusion of similar head parameters step by step, retaining the parametric knowledge of the MHA checkpoint. We construct DHA models by transforming various scales of MHA checkpoints given target head budgets. Our experiments show that DHA remarkably requires a mere 0.25\% of the original model's pre-training budgets to achieve 97.6\% of performance while saving 75\% of KV cache. Compared to Group-Query Attention (GQA), DHA achieves a 5$\times$ training acceleration, a maximum of 13.93\% performance improvement under 0.01\% pre-training budget, and 4\% relative improvement under 0.05\% pre-training budget.