Requests for name changes in the electronic proceedings will be accepted with no questions asked. However name changes may cause bibliographic tracking issues. Authors are asked to consider this carefully and discuss it with their co-authors prior to requesting a name change in the electronic proceedings. Use the Report an Issue link to request a name change.

The rapid scaling of large language models (LLMs) has significantly increased GPU memory pressure, which is further aggravated by training optimization techniques such as virtual pipeline and recomputation that disrupt tensor lifespans and introduce considerable memory fragmentation. Such fragmentation stems from the use of online GPU memory allocators in popular deep learning frameworks like PyTorch, which disregard tensor lifespans. As a result, this inefficiency can waste as much as 43% of memory and trigger out-of-memory errors, undermining the effectiveness of optimization methods. To address this, we introduce STAlloc, a GPU memory allocator for deep learning frameworks that reduces fragmentation by exploiting the spatial and temporal regularity in memory allocation behaviors of training workloads. STAlloc introduces a novel paradigm that combines offline planning with online allocation. The offline planning leverages spatio-temporal regularities to generate a near-optimal allocation plan, while the online allocation handles complex and dynamic models such as Mixture-of-Experts (MoE). Built as a pluggable PyTorch memory allocator, STAlloc reduces fragmentation ratio on average by 85.1% (up to 100%) across both dense and MoE models, with negligible overhead. This enables more efficient, high-throughput training configurations and improves throughput performance by up to 32.5%.

The growing memory footprint of the Key-Value (KV) cache poses a severe scalability bottleneck for long-context Large Language Model (LLM) inference. While KV cache eviction has emerged as an effective solution by discarding less critical tokens, existing token-, block-, and sentence-level compression methods struggle to balance semantic coherence and memory efficiency. To this end, we introduce SABlock, a \underline{s}emantic-aware KV cache eviction framework with \underline{a}daptive \underline{block} sizes. Specifically, SABlock first performs semantic segmentation to align compression boundaries with linguistic structures, then applies segment-guided token scoring to refine token importance estimation. Finally, for each segment, a budget-driven search strategy adaptively determines the optimal block size that preserves semantic integrity while improving compression efficiency under a given cache budget. Extensive experiments on long-context benchmarks demonstrate that SABlock consistently outperforms state-of-the-art baselines under the same memory budgets. For instance, on Needle-in-a-Haystack (NIAH), SABlock achieves 99.9% retrieval accuracy with only 96 KV entries, nearly matching the performance of the full-cache baseline that retains up to 8K entries. Under a fixed cache budget of 1,024, SABlock further reduces peak memory usage by 46.28% and achieves up to 9.5x faster decoding on a 128K context length.

Recent advances in Post-Training Quantization (PTQ) techniques have significantly increased demand for serving quantized large language models (LLMs), enabling higher throughput and substantially reduced memory usage with minimal accuracy loss. Quantized models address memory constraints in LLMs and enhance GPU resource utilization through efficient GPU sharing. However, quantized models have smaller KV block sizes than non-quantized models, causing limited memory efficiency due to memory fragmentation. Also, distinct resource usage patterns between quantized and non-quantized models require efficient scheduling to maximize throughput. To address these challenges, we propose FineServe, an inference serving framework for mixed-precision LLMs. FineServe's key contributions include: (1) KV Slab, a precision-aware adaptive memory management technique dynamically allocating KV cache based on model quantization characteristics, significantly reducing GPU memory fragmentation, and (2) a two-level scheduling framework comprising a global scheduler that places models to GPUs based on request rates, latency SLOs, and memory constraints and efficiency, and a local scheduler that adaptively adjusts batch sizes according to real-time request fluctuations. Experimental results demonstrate that FineServe achieves up to 2.2x higher SLO attainment and 1.8x higher token generation throughput compared to the state-of-the-art GPU sharing systems.

Large warehouse-scale computers (WSCs) underpin all the cloud computing services we use daily--whether it is Web search, video streaming, social networks, or even emerging AI chatbots or agents. The memory subsystem in these computers poses one of the biggest challenges in their design and operation: Across the industry, Big Tech companies such as Amazon, Google, Meta, and Microsoft spend billions of dollars buying memory and consume hundreds of megawatts powering them. Sadly, this problem is only getting worse, exacerbated by slowing of technology scaling trends (like Moore's law) and exploding demand for more data and correspondingly more memory--for example, artificial intelligence (AI) workloads. One approach to address the costs of memory is to use tiers. Most workloads have a working dataset that includes both hot (more frequently used) and cold (less frequently used) data.

Inspired by the Kolmogorov-Arnold representation theorem, KANs offer a novel framework for function approximation by replacing traditional neural network weights with learnable univariate functions. This design demonstrates significant potential as an efficient and interpretable alternative to traditional MLPs. However, KANs are characterized by a substantially larger number of trainable parameters, leading to challenges in memory efficiency and higher training costs compared to MLPs. To address this limitation, we propose to generate weights for KANs via a smaller meta-learner, called MetaKANs. By training KANs and MetaKANs in an end-to-end differentiable manner, MetaKANs achieve comparable or even superior performance while significantly reducing the number of trainable parameters and maintaining promising interpretability. Extensive experiments on diverse benchmark tasks, including symbolic regression, partial differential equation solving, and image classification, demonstrate the effectiveness of MetaKANs in improving parameter efficiency and memory usage. The proposed method provides an alternative technique for training KANs, that allows for greater scalability and extensibility, and narrows the training cost gap with MLPs stated in the original paper of KANs. Our code is available at https://github.com/Murphyzc/MetaKAN.

While long-context large language models (LLMs) exhibit remarkable document processing capabilities, their prohibitively high training costs often hinder customized applications. To mitigate this issue, we propose \textit{Sequential Chunk-wise Optimization} (SeCO), a memory-efficient training paradigm that partitions lengthy inputs into manageable chunks. Each chunk independently constructs its computational graph and performs localized backpropagation, ensuring that only one chunk's forward activations are stored in memory. Building on SeCO, we further introduce \textit{Sparse Chunk-wise Optimization} (SpaCO), which reduces computational overhead by selectively propagating gradients to specific chunks and incorporates a carefully designed compensation factor to ensure unbiased gradient estimation. SpaCO decouples the computational cost of backpropagation from the context length, enabling training time to gradually converge to inference time as sequences become longer. Implemented as lightweight training wrappers, both SeCO and SpaCO offer substantial practical benefits. For example, when fine-tuning an 8B model with LoRA on a single RTX 3090 GPU, SeCO expands maximum sequence length from 1K to 16K tokens, while SpaCO demonstrates accelerated training speed -- achieving up to 3x faster than SeCO under the same experimental setup. These innovations provide new insights into optimizing long-context models, making them more accessible for practical applications. We have open-sourced the code at \href{https://github.com/wenhaoli-xmu/seco}{here}.

Building upon the success of low-rank adapter (LoRA), low-rank gradient projection (LoRP) has emerged as a promising solution for memory-efficient fine-tuning. However, existing LoRP methods typically treat each row of the gradient matrix as the default projection unit, leaving the role of projection granularity underexplored. In this work, we propose a novel framework, VLoRP, that extends low-rank gradient projection by introducing an additional degree of freedom for controlling the trade-off between memory efficiency and performance, beyond the rank hyper-parameter. Through this framework, we systematically explore the impact of projection granularity, demonstrating that finer-grained projections lead to enhanced stability and efficiency even under a fixed memory budget. Regarding the optimization for VLoRP, we present ProjFactor, an adaptive memory-efficient optimizer, that significantly reduces memory requirement while ensuring competitive performance, even in the presence of gradient accumulation. Additionally, we provide a theoretical analysis of VLoRP, demonstrating the descent and convergence of its optimization trajectory under both SGD and ProjFactor. Extensive experiments are conducted to validate our findings, covering tasks such as commonsense reasoning, MMLU, and GSM8K.

Diffusion models have recently gained state of the art performance on many image generation tasks. However, most models require significant computational resources to achieve this. This becomes apparent in the application of medical image synthesis due to the 3D nature of medical datasets like CT-scans, MRIs, electron microscope, etc. In this paper we propose a novel architecture for a single GPU memory-efficient training for diffusion models for high dimensional medical datasets. The proposed model is built by using an invertible UNet architecture with invertible attention modules. This leads to the following two contributions: 1. denoising diffusion models and thus enabling memory usage to be independent of the dimensionality of the dataset, and 2. reducing the energy usage during training. While this new model can be applied to a multitude of image generation tasks, we showcase its memory-efficiency on the 3D BraTS2020 dataset leading to up to 15\% decrease in peak memory consumption during training with comparable results to SOTA while maintaining the image quality.