#1
arXivJune 24, 2025

Tensor-Parallelism with Partially Synchronized Activations

By:Itay Lamprecht, Asaf Karnieli, Yair Hanani, Niv Giladi, Daniel Soudry

Training and inference of Large Language Models (LLMs) with tensor-parallelism requires substantial communication to synchronize activations. Our findings suggest that with a few minor adjustments to current practices, LLMs can be trained without fully synchronizing activations, reducing bandwidth demands. We name this "Communication-Aware Architecture for Tensor-parallelism" (CAAT-Net). We train 1B and 7B parameter CAAT-Net models, with a 50% reduction in tensor-parallel communication and no significant drop in pretraining accuracy. Furthermore, we demonstrate how CAAT-Net accelerates both training and inference workloads.

Explore summaries→
#2
arXivJuly 17, 2025

TransEvalnia: Reasoning-based Evaluation and Ranking of Translations

By:Richard Sproat, Tianyu Zhao, Llion Jones

We present TransEvalnia, a prompting-based translation evaluation and ranking system that uses reasoning in performing its evaluations and ranking. This system presents fine-grained evaluations based on a subset of the Multidimensional Quality Metrics (https://themqm.org/), returns an assessment of which translation it deems the best, and provides numerical scores for the various dimensions and for the overall translation. We show that TransEvalnia performs as well as or better than the state-of-the-art MT-Ranker (Moosa et al. 2024) on our own English-Japanese data as well as several language pairs from various WMT shared tasks. Using Anthropic's Claude-3.5-Sonnet and Qwen-2.5-72B-Instruct as the evaluation LLMs, we show that the evaluations returned are deemed highly acceptable to human raters, and that the scores assigned to the translations by Sonnet, as well as other LLMs, correlate well with scores assigned by the human raters. We also note the sensitivity of our system -- as well as MT-Ranker -- to the order in which the translations are presented, and we propose methods to address this position bias. All data, including the system's evaluation and reasoning, human assessments, as well as code is released.

Explore summaries→
#3
arXivAugust 15, 2023

Attention Is Not All You Need Anymore

By:Zhe Chen

In recent years, the popular Transformer architecture has achieved great success in many application areas, including natural language processing and computer vision. Many existing works aim to reduce the computational and memory complexity of the self-attention mechanism in the Transformer by trading off performance. However, performance is key for the continuing success of the Transformer. In this paper, a family of drop-in replacements for the self-attention mechanism in the Transformer, called the Extractors, is proposed. Four types of the Extractors, namely the super high-performance Extractor (SHE), the higher-performance Extractor (HE), the worthwhile Extractor (WE), and the minimalist Extractor (ME), are proposed as examples. Experimental results show that replacing the self-attention mechanism with the SHE evidently improves the performance of the Transformer, whereas the simplified versions of the SHE, i.e., the HE, the WE, and the ME, perform close to or better than the self-attention mechanism with less computational and memory complexity. Furthermore, the proposed Extractors have the potential or are able to run faster than the self-attention mechanism since their critical paths of computation are much shorter. Additionally, the sequence prediction problem in the context of text generation is formulated using variable-length discrete-time Markov chains, and the Transformer is reviewed based on our understanding.

Explore summaries→
#4
arXivJune 12, 2017

Attention Is All You Need

By:Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, Illia Polosukhin

The dominant sequence transduction models are based on complex recurrent or convolutional neural networks in an encoder-decoder configuration. The best performing models also connect the encoder and decoder through an attention mechanism. We propose a new simple network architecture, the Transformer, based solely on attention mechanisms, dispensing with recurrence and convolutions entirely. Experiments on two machine translation tasks show these models to be superior in quality while being more parallelizable and requiring significantly less time to train. Our model achieves 28.4 BLEU on the WMT 2014 English-to-German translation task, improving over the existing best results, including ensembles by over 2 BLEU. On the WMT 2014 English-to-French translation task, our model establishes a new single-model state-of-the-art BLEU score of 41.8 after training for 3.5 days on eight GPUs, a small fraction of the training costs of the best models from the literature. We show that the Transformer generalizes well to other tasks by applying it successfully to English constituency parsing both with large and limited training data.

Explore summaries→
#5
arXivDecember 10, 2015

Deep Residual Learning for Image Recognition

By:Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun

Deeper neural networks are more difficult to train. We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. We explicitly reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions. We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth. On the ImageNet dataset we evaluate residual nets with a depth of up to 152 layers---8x deeper than VGG nets but still having lower complexity. An ensemble of these residual nets achieves 3.57% error on the ImageNet test set. This result won the 1st place on the ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100 and 1000 layers. The depth of representations is of central importance for many visual recognition tasks. Solely due to our extremely deep representations, we obtain a 28% relative improvement on the COCO object detection dataset. Deep residual nets are foundations of our submissions to ILSVRC & COCO 2015 competitions, where we also won the 1st places on the tasks of ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation.

Explore summaries→
#6
arXivJune 1, 2023

AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration

By:Ji Lin, Jiaming Tang, Haotian Tang, Shang Yang, Wei-Ming Chen, Wei-Chen Wang, Guangxuan Xiao, Xingyu Dang, Chuang Gan, Song Han

Large language models (LLMs) have transformed numerous AI applications. On-device LLM is becoming increasingly important: running LLMs locally on edge devices can reduce the cloud computing cost and protect users' privacy. However, the astronomical model size and the limited hardware resource pose significant deployment challenges. We propose Activation-aware Weight Quantization (AWQ), a hardware-friendly approach for LLM low-bit weight-only quantization. AWQ finds that not all weights in an LLM are equally important. Protecting only 1% salient weights can greatly reduce quantization error. To identify salient weight channels, we should refer to the activation distribution, not weights. To avoid the hardware-inefficient mix-precision quantization, we mathematically derive that scaling up the salient channels can reduce the quantization error. AWQ employs an equivalent transformation to scale the salient weight channels to protect them. The scale is determined by collecting the activation statistics offline. AWQ does not rely on any backpropagation or reconstruction, so it generalizes to different domains and modalities without overfitting the calibration set. AWQ outperforms existing work on various language modeling and domain-specific benchmarks (coding and math). Thanks to better generalization, it achieves excellent quantization performance for instruction-tuned LMs and, for the first time, multi-modal LMs. Alongside AWQ, we implement TinyChat, an efficient and flexible inference framework tailored for 4-bit on-device LLM/VLMs. With kernel fusion and platform-aware weight packing, TinyChat offers more than 3x speedup over the Huggingface FP16 implementation on both desktop and mobile GPUs. It also democratizes the deployment of the 70B Llama-2 model on mobile GPUs.

Explore summaries→
#7
arXivFebruary 23, 2025

Compression Scaling Laws:Unifying Sparsity and Quantization

By:Elias Frantar, Utku Evci, Wonpyo Park, Neil Houlsby, Dan Alistarh

We investigate how different compression techniques -- such as weight and activation quantization, and weight sparsity -- affect the scaling behavior of large language models (LLMs) during pretraining. Building on previous work showing that weight sparsity acts as a constant multiplier on model size in scaling laws, we demonstrate that this "effective parameter" scaling pattern extends to quantization as well. Specifically, we establish that weight-only quantization achieves strong parameter efficiency multipliers, while full quantization of both weights and activations shows diminishing returns at lower bitwidths. Our results suggest that different compression techniques can be unified under a common scaling law framework, enabling principled comparison and combination of these methods.

Explore summaries→
#8
arXivOctober 31, 2022

GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers

By:Elias Frantar, Saleh Ashkboos, Torsten Hoefler, Dan Alistarh

Generative Pre-trained Transformer models, known as GPT or OPT, set themselves apart through breakthrough performance across complex language modelling tasks, but also by their extremely high computational and storage costs. Specifically, due to their massive size, even inference for large, highly-accurate GPT models may require multiple performant GPUs, which limits the usability of such models. While there is emerging work on relieving this pressure via model compression, the applicability and performance of existing compression techniques is limited by the scale and complexity of GPT models. In this paper, we address this challenge, and propose GPTQ, a new one-shot weight quantization method based on approximate second-order information, that is both highly-accurate and highly-efficient. Specifically, GPTQ can quantize GPT models with 175 billion parameters in approximately four GPU hours, reducing the bitwidth down to 3 or 4 bits per weight, with negligible accuracy degradation relative to the uncompressed baseline. Our method more than doubles the compression gains relative to previously-proposed one-shot quantization methods, preserving accuracy, allowing us for the first time to execute an 175 billion-parameter model inside a single GPU for generative inference. Moreover, we also show that our method can still provide reasonable accuracy in the extreme quantization regime, in which weights are quantized to 2-bit or even ternary quantization levels. We show experimentally that these improvements can be leveraged for end-to-end inference speedups over FP16, of around 3.25x when using high-end GPUs (NVIDIA A100) and 4.5x when using more cost-effective ones (NVIDIA A6000). The implementation is available at https://github.com/IST-DASLab/gptq.

Explore summaries→
#9
arXivAugust 13, 2025

Noise Hypernetworks: Amortizing Test-Time Compute in Diffusion Models

By:Luca Eyring, Shyamgopal Karthik, Alexey Dosovitskiy, Nataniel Ruiz, Zeynep Akata

The new paradigm of test-time scaling has yielded remarkable breakthroughs in Large Language Models (LLMs) (e.g. reasoning models) and in generative vision models, allowing models to allocate additional computation during inference to effectively tackle increasingly complex problems. Despite the improvements of this approach, an important limitation emerges: the substantial increase in computation time makes the process slow and impractical for many applications. Given the success of this paradigm and its growing usage, we seek to preserve its benefits while eschewing the inference overhead. In this work we propose one solution to the critical problem of integrating test-time scaling knowledge into a model during post-training. Specifically, we replace reward guided test-time noise optimization in diffusion models with a Noise Hypernetwork that modulates initial input noise. We propose a theoretically grounded framework for learning this reward-tilted distribution for distilled generators, through a tractable noise-space objective that maintains fidelity to the base model while optimizing for desired characteristics. We show that our approach recovers a substantial portion of the quality gains from explicit test-time optimization at a fraction of the computational cost. Code is available at https://github.com/ExplainableML/HyperNoise

Explore summaries→
#10
arXivOctober 22, 2020

An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

By:Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby

While the Transformer architecture has become the de-facto standard for natural language processing tasks, its applications to computer vision remain limited. In vision, attention is either applied in conjunction with convolutional networks, or used to replace certain components of convolutional networks while keeping their overall structure in place. We show that this reliance on CNNs is not necessary and a pure transformer applied directly to sequences of image patches can perform very well on image classification tasks. When pre-trained on large amounts of data and transferred to multiple mid-sized or small image recognition benchmarks (ImageNet, CIFAR-100, VTAB, etc.), Vision Transformer (ViT) attains excellent results compared to state-of-the-art convolutional networks while requiring substantially fewer computational resources to train.

Explore summaries→
#11
arXivOctober 1, 2018

Tame topology of arithmetic quotients and algebraicity of Hodge loci

By:Benjamin Bakker, Bruno Klingler, Jacob Tsimerman

In this paper we prove the following results: $1)$ We show that any arithmetic quotient of a homogeneous space admits a natural real semi-algebraic structure for which its Hecke correspondences are semi-algebraic. A particularly important example is given by Hodge varieties, which parametrize pure polarized integral Hodge structures. $2)$ We prove that the period map associated to any pure polarized variation of integral Hodge structures $\mathbb{V}$ on a smooth complex quasi-projective variety $S$ is definable with respect to an o-minimal structure on the relevant Hodge variety induced by the above semi-algebraic structure. $3)$ As a corollary of $2)$ and of Peterzil-Starchenko's o-minimal Chow theorem we recover that the Hodge locus of $(S, \mathbb{V})$ is a countable union of algebraic subvarieties of $S$, a result originally due to Cattani-Deligne-Kaplan. Our approach simplifies the proof of Cattani-Deligne-Kaplan, as it does not use the full power of the difficult multivariable $SL_2$-orbit theorem of Cattani-Kaplan-Schmid.

Explore summaries→
#12
arXivOctober 9, 2018

On the Performance of DF-based Power-Line/Visible-Light Communication Systems

By:Waled Gheth, Khaled M. Rabie, Bamidele Adebisi, Muhammad Ijaz, Georgina Harris

This paper presents a comprehensive performance analysis of an integrated indoor power line communication (PLC)/visible light communication (VLC) system with the presence of a decode-and-forward (DF) relay. The existing indoor power line networks are used as the backbone for VLCs. The performance of the proposed system is evaluated in terms of the average capacity and the outage probability. A new unified mathematical method is developed for the PLC/VLC system and analytical expressions for the aforementioned performance metrics are derived. Monte Carlo simulations are provided throughout the paper to verify the correctness of the analysis. The results reveal that the performance of the proposed system deteriorates with increasing the end-to-end distance and improves with increasing the relay transmit power. It is also shown that the outage probability of the system under consideration is negatively affected by the vertical distance to user plane.

Explore summaries→