Low-rank adaptation (LoRA) methods show great potential for scaling pre-trained general-purpose Large Language Models (LLMs) to hundreds or thousands of use scenarios. However, their efficacy in high-stakes domains like finance is rarely explored, e.g., passing CFA exams and analyzing SEC filings. In this paper, we present the open-source FinLoRA project that benchmarks LoRA methods on both general and highly professional financial tasks. First, we curated 19 datasets covering diverse financial applications; in particular, we created four novel XBRL analysis datasets based on 150 SEC filings. Second, we evaluated five LoRA methods and five base LLMs. Finally, we provide extensive experimental results in terms of accuracy, F1, and BERTScore and report computational cost in terms of time and GPU memory during fine-tuning and inference stages. We find that LoRA methods achieved substantial performance gains of 36\% on average over base models. Our FinLoRA project provides an affordable and scalable approach to democratize financial intelligence to the general public. Datasets, LoRA adapters, code, and documentation are available at https://github.com/Open-Finance-Lab/FinLoRA
Foundation models are increasingly becoming better autonomous programmers, raising the prospect that they could also automate dangerous offensive cyber-operations. Current frontier model audits probe the cybersecurity risks of such agents, but most fail to account for the degrees of freedom available to adversaries in the real world. In particular, with strong verifiers and financial incentives, agents for offensive cybersecurity are amenable to iterative improvement by would-be adversaries. We argue that assessments should take into account an expanded threat model in the context of cybersecurity, emphasizing the varying degrees of freedom that an adversary may possess in stateful and non-stateful environments within a fixed compute budget. We show that even with a relatively small compute budget (8 H100 GPU Hours in our study), adversaries can improve an agent's cybersecurity capability on InterCode CTF by more than 40\% relative to the baseline -- without any external assistance. These results highlight the need to evaluate agents' cybersecurity risk in a dynamic manner, painting a more representative picture of risk.
Drug combinations play a critical role in cancer therapy by significantly enhancing treatment efficacy and overcoming drug resistance. However, the combinatorial space of possible drug pairs grows exponentially, making experimental screening highly impractical. Therefore, developing efficient computational methods to predict promising drug combinations and guide experimental validation is of paramount importance. In this work, we propose ADGSyn, an innovative method for predicting drug synergy. The key components of our approach include: (1) shared projection matrices combined with attention mechanisms to enable cross-drug feature alignment; (2) automatic mixed precision (AMP)-optimized graph operations that reduce memory consumption by 40\% while accelerating training speed threefold; and (3) residual pathways stabilized by LayerNorm to ensure stable gradient propagation during training. Evaluated on the O'Neil dataset containing 13,243 drug--cell line combinations, ADGSyn demonstrates superior performance over eight baseline methods. Moreover, the framework supports full-batch processing of up to 256 molecular graphs on a single GPU, setting a new standard for efficiency in drug synergy prediction within the field of computational oncology.
Visual autoregressive (VAR) modeling has marked a paradigm shift in image generation from next-token prediction to next-scale prediction. VAR predicts a set of tokens at each step from coarse to fine scale, leading to better image quality and faster inference speed compared to existing diffusion models. However, the large parameter size and computation cost hinder its deployment on edge devices. To reduce the memory and computation cost, we propose FPQVAR, an efficient post-training floating-point (FP) quantization framework for VAR featuring algorithm and hardware co-design. At the algorithm level, we first identify the challenges of quantizing VAR. To address them, we propose Dual Format Quantization for the highly imbalanced input activation. We further propose Group-wise Hadamard Transformation and GHT-Aware Learnable Transformation to address the time-varying outlier channels. At the hardware level, we design the first low-bit FP quantizer and multiplier with lookup tables on FPGA and propose the first FPGA-based VAR accelerator featuring low-bit FP computation and an elaborate two-level pipeline. Extensive experiments show that compared to the state-of-the-art quantization method, our proposed FPQVAR significantly improves Fr\'echet Inception Distance (FID) from 10.83 to 3.58, Inception Score (IS) from 175.9 to 241.5 under 4-bit quantization. FPQVAR also significantly improves the performance of 6-bit quantized VAR, bringing it on par with the FP16 model. Our accelerator on AMD-Xilinx VCK190 FPGA achieves a throughput of 1.1 image/s, which is 3.1x higher than the integer-based accelerator. It also demonstrates 3.6x and 2.8x higher energy efficiency compared to the integer-based accelerator and GPU baseline, respectively.
Long-video understanding has emerged as a crucial capability in real-world applications such as video surveillance, meeting summarization, educational lecture analysis, and sports broadcasting. However, it remains computationally prohibitive for VideoLLMs, primarily due to two bottlenecks: 1) sequential video decoding, the process of converting the raw bit stream to RGB frames can take up to a minute for hour-long video inputs, and 2) costly prefilling of up to several million tokens for LLM inference, resulting in high latency and memory use. To address these challenges, we propose QuickVideo, a system-algorithm co-design that substantially accelerates long-video understanding to support real-time downstream applications. It comprises three key innovations: QuickDecoder, a parallelized CPU-based video decoder that achieves 2-3 times speedup by splitting videos into keyframe-aligned intervals processed concurrently; QuickPrefill, a memory-efficient prefilling method using KV-cache pruning to support more frames with less GPU memory; and an overlapping scheme that overlaps CPU video decoding with GPU inference. Together, these components infernece time reduce by a minute on long video inputs, enabling scalable, high-quality video understanding even on limited hardware. Experiments show that QuickVideo generalizes across durations and sampling rates, making long video processing feasible in practice.
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}.
Despite decades of advancements, the simulation of fluids remains one of the most challenging areas of in scientific computing. Supported by the necessity of gradient information in deep learning, differentiable simulators have emerged as an effective tool for optimization and learning in physics simulations. In this work, we present our fluid simulator PICT, a differentiable pressure-implicit solver coded in PyTorch with Graphics-processing-unit (GPU) support. We first verify the accuracy of both the forward simulation and our derived gradients in various established benchmarks like lid-driven cavities and turbulent channel flows before we show that the gradients provided by our solver can be used to learn complicated turbulence models in 2D and 3D. We apply both supervised and unsupervised training regimes using physical priors to match flow statistics. In particular, we learn a stable sub-grid scale (SGS) model for a 3D turbulent channel flow purely based on reference statistics. The low-resolution corrector trained with our solver runs substantially faster than the highly resolved references, while keeping or even surpassing their accuracy. Finally, we give additional insights into the physical interpretation of different solver gradients, and motivate a physically informed regularization technique. To ensure that the full potential of PICT can be leveraged, it is published as open source: https://github.com/tum-pbs/PICT.
We introduce CASS, the first large-scale dataset and model suite for cross-architecture GPU code transpilation, targeting both source-level (CUDA <--> HIP) and assembly-level (Nvidia SASS <--> AMD RDNA3) translation. The dataset comprises 70k verified code pairs across host and device, addressing a critical gap in low-level GPU code portability. Leveraging this resource, we train the CASS family of domain-specific language models, achieving 95% source translation accuracy and 37.5% assembly translation accuracy, substantially outperforming commercial baselines such as GPT-4o, Claude, and Hipify. Our generated code matches native performance in over 85% of test cases, preserving runtime and memory behavior. To support rigorous evaluation, we introduce CASS-Bench, a curated benchmark spanning 16 GPU domains with ground-truth execution. All data, models, and evaluation tools are released as open source to foster progress in GPU compiler tooling, binary compatibility, and LLM-guided hardware translation.
Large language models (LLMs) have made exciting achievements across various domains, yet their deployment on resource-constrained personal devices remains hindered by the prohibitive computational and memory demands of task-specific fine-tuning. While quantization offers a pathway to efficiency, existing methods struggle to balance performance and overhead, either incurring high computational/memory costs or failing to address activation outliers, a critical bottleneck in quantized fine-tuning. To address these challenges, we propose the Outlier Spatial Stability Hypothesis (OSSH): During fine-tuning, certain activation outlier channels retain stable spatial positions across training iterations. Building on OSSH, we propose Quaff, a Quantized parameter-efficient fine-tuning framework for LLMs, optimizing low-precision activation representations through targeted momentum scaling. Quaff dynamically suppresses outliers exclusively in invariant channels using lightweight operations, eliminating full-precision weight storage and global rescaling while reducing quantization errors. Extensive experiments across ten benchmarks validate OSSH and demonstrate Quaff's efficacy. Specifically, on the GPQA reasoning benchmark, Quaff achieves a 1.73x latency reduction and 30% memory savings over full-precision fine-tuning while improving accuracy by 0.6% on the Phi-3 model, reconciling the triple trade-off between efficiency, performance, and deployability. By enabling consumer-grade GPU fine-tuning (e.g., RTX 2080 Super) without sacrificing model utility, Quaff democratizes personalized LLM deployment. The code is available at https://github.com/Little0o0/Quaff.git.
Low-rank regularization (LRR) has been widely applied in various machine learning tasks, but the associated optimization is challenging. Directly optimizing the rank function under constraints is NP-hard in general. To overcome this difficulty, various relaxations of the rank function were studied. However, optimization of these relaxed LRRs typically depends on singular value decomposition, which is a time-consuming and nondifferentiable operator that cannot be optimized with gradient-based techniques. To address these challenges, in this paper we propose an efficient differentiable approximation of the generalized LRR. The considered LRR form subsumes many popular choices like the nuclear norm, the Schatten-$p$ norm, and various nonconvex relaxations. Our method enables LRR terms to be appended to loss functions in a plug-and-play fashion, and the GPU-friendly operations enable efficient and convenient implementation. Furthermore, convergence analysis is presented, which rigorously shows that both the bias and the variance of our rank estimator rapidly reduce with increased sample size and iteration steps. In the experimental study, the proposed method is applied to various tasks, which demonstrates its versatility and efficiency. Code is available at https://github.com/naiqili/EDLRR.
Accelerating large language model (LLM) inference is critical for real-world deployments requiring high throughput and low latency. Contextual sparsity, where each token dynamically activates only a small subset of the model parameters, shows promise but does not scale to large batch sizes due to union of active neurons quickly approaching dense computation. We introduce Polar Sparsity, highlighting a key shift in sparsity importance from MLP to Attention layers as we scale batch size and sequence length. While MLP layers become more compute-efficient under batching, their sparsity vanishes. In contrast, attention becomes increasingly more expensive at scale, while their head sparsity remains stable and batch-invariant. We develop hardware-efficient, sparsity-aware GPU kernels for selective MLP and Attention computations, delivering up to \(2.2\times\) end-to-end speedups for models like OPT, LLaMA-2 \& 3, across various batch sizes and sequence lengths without compromising accuracy. To our knowledge, this is the first work to demonstrate that contextual sparsity can scale effectively to large batch sizes, delivering substantial inference acceleration with minimal changes, making Polar Sparsity practical for large-scale, high-throughput LLM deployment systems. Our code is available at: https://github.com/susavlsh10/Polar-Sparsity.
While Transformer self-attention offers strong parallelism, the Key-Value (KV) cache grows linearly with sequence length and becomes a bottleneck for inference efficiency. Multi-head latent attention was recently developed to compress the KV cache into a low-rank latent space. This paper proposes Multi-head Temporal Latent Attention (MTLA), which further reduces the KV cache size along the temporal dimension, greatly lowering the memory footprint of self-attention inference. MTLA employs a hyper-network to dynamically merge temporally adjacent KV cache vectors. To address the mismatch between the compressed KV cache and processed sequence lengths, a stride-aware causal mask is proposed to ensure efficient parallel training and consistency with inference behaviour. Experiments across tasks, including speech translation, speech recognition, speech understanding and text summarisation, demonstrate that MTLA achieves competitive performance compared to standard Multi-Head Attention (MHA), while greatly improving inference speed and GPU memory usage. For example, on a English-German speech translation task, MTLA achieves a 5.3x speedup and a reduction in GPU memory usage by a factor of 8.3 compared to MHA, while maintaining translation quality.
Lifelong learning enables large language models (LLMs) to adapt to evolving information by continually updating their internal knowledge. An ideal system should support efficient, wide-ranging updates while preserving existing capabilities and ensuring reliable deployment. Model editing stands out as a promising solution for this goal, offering a focused and efficient way to revise a model's internal knowledge. Although recent paradigms have made notable progress, they often struggle to meet the demands of practical lifelong adaptation at scale. To bridge this gap, we propose ULTRAEDIT-a fundamentally new editing solution that is training-, subject- and memory-free, making it particularly well-suited for ultra-scalable, real-world lifelong model editing. ULTRAEDIT performs editing through a self-contained process that relies solely on lightweight linear algebra operations to compute parameter shifts, enabling fast and consistent parameter modifications with minimal overhead. To improve scalability in lifelong settings, ULTRAEDIT employs a lifelong normalization strategy that continuously updates feature statistics across turns, allowing it to adapt to distributional shifts and maintain consistency over time. ULTRAEDIT achieves editing speeds over 7x faster than the previous state-of-the-art method-which was also the fastest known approach-while consuming less than 1/3 the VRAM, making it the only method currently capable of editing a 7B LLM on a 24GB consumer-grade GPU. Furthermore, we construct ULTRAEDITBENCH-the largest dataset in the field to date, with over 2M editing pairs-and demonstrate that our method supports up to 1M edits while maintaining high accuracy. Comprehensive experiments on four datasets and six models show that ULTRAEDIT consistently achieves superior performance across diverse model editing scenarios. Our code is available at: https://github.com/XiaojieGu/UltraEdit.
Diffusion-based Transformers have demonstrated impressive generative capabilities, but their high computational costs hinder practical deployment, for example, generating an $8192\times 8192$ image can take over an hour on an A100 GPU. In this work, we propose GRAT (\textbf{GR}ouping first, \textbf{AT}tending smartly), a training-free attention acceleration strategy for fast image and video generation without compromising output quality. The key insight is to exploit the inherent sparsity in learned attention maps (which tend to be locally focused) in pretrained Diffusion Transformers and leverage better GPU parallelism. Specifically, GRAT first partitions contiguous tokens into non-overlapping groups, aligning both with GPU execution patterns and the local attention structures learned in pretrained generative Transformers. It then accelerates attention by having all query tokens within the same group share a common set of attendable key and value tokens. These key and value tokens are further restricted to structured regions, such as surrounding blocks or criss-cross regions, significantly reducing computational overhead (e.g., attaining a \textbf{35.8$\times$} speedup over full attention when generating $8192\times 8192$ images) while preserving essential attention patterns and long-range context. We validate GRAT on pretrained Flux and HunyuanVideo for image and video generation, respectively. In both cases, GRAT achieves substantially faster inference without any fine-tuning, while maintaining the performance of full attention. We hope GRAT will inspire future research on accelerating Diffusion Transformers for scalable visual generation.
Invariant Point Attention (IPA) is a key algorithm for geometry-aware modeling in structural biology, central to many protein and RNA models. However, its quadratic complexity limits the input sequence length. We introduce FlashIPA, a factorized reformulation of IPA that leverages hardware-efficient FlashAttention to achieve linear scaling in GPU memory and wall-clock time with sequence length. FlashIPA matches or exceeds standard IPA performance while substantially reducing computational costs. FlashIPA extends training to previously unattainable lengths, and we demonstrate this by re-training generative models without length restrictions and generating structures of thousands of residues. FlashIPA is available at https://github.com/flagshippioneering/flash_ipa.
As the size of large language models grows exponentially, GPU memory has become a bottleneck for adapting these models to downstream tasks. In this paper, we aim to push the limits of memory-efficient training by minimizing memory usage on model weights, gradients, and optimizer states, within a unified framework. Our idea is to eliminate both gradients and optimizer states using zeroth-order optimization, which approximates gradients by perturbing weights during forward passes to identify gradient directions. To minimize memory usage on weights, we employ model quantization, e.g., converting from bfloat16 to int4. However, directly applying zeroth-order optimization to quantized weights is infeasible due to the precision gap between discrete weights and continuous gradients, which would otherwise require de-quantization and re-quantization. To overcome this challenge, we propose Quantized Zeroth-order Optimization (QZO), a novel approach that perturbs the continuous quantization scale for gradient estimation and uses a directional derivative clipping method to stabilize training. QZO is orthogonal to both scalar-based and codebook-based post-training quantization methods. Compared to full-parameter fine-tuning in bfloat16, QZO can reduce the total memory cost by more than 18$\times$ for 4-bit LLMs, and enables fine-tuning Llama-2-13B and Stable Diffusion 3.5 Large within a single 24GB GPU.
Essential to visual generation is efficient modeling of visual data priors. Conventional next-token prediction methods define the process as learning the conditional probability distribution of successive tokens. Recently, next-scale prediction methods redefine the process to learn the distribution over multi-scale representations, significantly reducing generation latency. However, these methods condition each scale on all previous scales and require each token to consider all preceding tokens, exhibiting scale and spatial redundancy. To better model the distribution by mitigating redundancy, we propose Markovian Visual AutoRegressive modeling (MVAR), a novel autoregressive framework that introduces scale and spatial Markov assumptions to reduce the complexity of conditional probability modeling. Specifically, we introduce a scale-Markov trajectory that only takes as input the features of adjacent preceding scale for next-scale prediction, enabling the adoption of a parallel training strategy that significantly reduces GPU memory consumption. Furthermore, we propose spatial-Markov attention, which restricts the attention of each token to a localized neighborhood of size k at corresponding positions on adjacent scales, rather than attending to every token across these scales, for the pursuit of reduced modeling complexity. Building on these improvements, we reduce the computational complexity of attention calculation from O(N^2) to O(Nk), enabling training with just eight NVIDIA RTX 4090 GPUs and eliminating the need for KV cache during inference. Extensive experiments on ImageNet demonstrate that MVAR achieves comparable or superior performance with both small model trained from scratch and large fine-tuned models, while reducing the average GPU memory footprint by 3.0x.
Large Language Models (LLMs)-based hybrid retrieval uses LLMs to encode queries and documents into low-dimensional dense or high-dimensional sparse vectors. It retrieves documents relevant to search queries based on vector similarities. Documents are pre-encoded offline, while queries arrive in real-time, necessitating an efficient online query encoder. Although LLMs significantly enhance retrieval capabilities, serving deeply parameterized LLMs slows down query inference throughput and increases demands for online deployment resources. In this paper, we propose LightRetriever, a novel LLM-based hybrid retriever with extremely lightweight query encoders. Our method retains a full-sized LLM for document encoding, but reduces the workload of query encoding to no more than an embedding lookup. Compared to serving a full-sized LLM on an H800 GPU, our approach achieves over a 1000x speedup for query inference with GPU acceleration, and even a 20x speedup without GPU. Experiments on large-scale retrieval benchmarks demonstrate that our method generalizes well across diverse retrieval tasks, retaining an average of 95% full-sized performance.
Tiny QA Benchmark++ (TQB++) presents an ultra-lightweight, multilingual smoke-test suite designed to give large-language-model (LLM) pipelines a unit-test style safety net dataset that runs in seconds with minimal cost. Born out of the tight feedback-loop demands building the Comet Opik prompt-optimization SDK, where waiting on heavyweight benchmarks breaks developer flow. TQB++ couples a 52-item English gold set (less than 20 kB) with a tiny synthetic-data generator pypi package built on provider-agnostic LiteLLM. The generator lets practitioners mint their own tiny packs in any language, domain, or difficulty, while ten ready-made packs already cover Arabic, Chinese, French, German, Japanese, Korean, Portuguese, Russian, Spanish, and Turkish. Every dataset ships with Croissant metadata and plug-and-play files for OpenAI-Evals, LangChain, and standard CI tools, so teams can drop deterministic micro-benchmarks directly into pull-request gates, prompt-engineering loops, and production dashboards without touching GPU budgets. A complete TQB++ run adds only a few seconds to pipeline latency yet reliably flags prompt-template errors, tokenizer drift, and fine-tuning side-effects long before full-scale suites like MMLU or BIG-Bench would finish configuring. The entire framework is released to accelerate continuous, resource-efficient quality assurance across the generative-AI ecosystem.
Modern gradient-boosted decision trees (GBDTs) accelerate split finding with histogram-based binning, which reduces complexity from O(N) to O(B) given a fixed bin budget B. However, the predominant quantile binning strategy-designed to distribute data points evenly among bins-may overlook critical boundary values that could enhance predictive performance. In this work, we propose replacing quantile binning with a k-means discretizer initialized with quantile bins. We test this swap on 33 OpenML tasks plus synthetics that control for modality, skew, and bin budget. Across 18 regression datasets, k-means shows no statistically significant losses at the 5% level and wins in four cases-most strikingly a 55% MSE drop on one particularly skewed dataset-even though k-means' mean reciprocal rank (MRR) is slightly lower (0.65 vs 0.72). On the 15 classification datasets the two methods are statistically tied (MRR 0.70 vs 0.68) with gaps $\leq$0.2 pp. Synthetic experiments confirm consistently large MSE gains-typically >20% and rising to 90% as outlier magnitude increases or bin budget drops. We find that k-means keeps error on par with exact splitting when extra cuts add little value, yet still recovers key split points that quantile overlooks. As such, we advocate for a built-in bin_method=k-means flag, especially in regression tasks and in tight-budget settings such as the 32-64-bin GPU regime-because it is a "safe default" with large upside, yet adds only a one-off, cacheable overhead ($\approx$ 2s to bin 10M rows on one core).