June Yong Yang

Jung Hyun Lee, Jeonghoon Kim, June Yong Yang et al.

With the commercialization of large language models (LLMs), weight-activation quantization has emerged to compress and accelerate LLMs, achieving high throughput while reducing inference costs. However, existing post-training quantization (PTQ) techniques for quantizing weights and activations of LLMs still suffer from non-negligible accuracy drops, especially on massive multitask language understanding. To address this issue, we propose Low-Rank Quantization (LRQ) - a simple yet effective post-training weight quantization method for LLMs that reconstructs the outputs of an intermediate Transformer block by leveraging low-rank weight-scaling matrices, replacing the conventional full weight-scaling matrices that entail as many learnable scales as their associated weights. Thanks to parameter sharing via low-rank structure, LRQ only needs to learn significantly fewer parameters while enabling the individual scaling of weights, thus boosting the generalization capability of quantized LLMs. We show the superiority of LRQ over prior LLM PTQ works under (i) 8-bit weight and per-tensor activation quantization, (ii) 4-bit weight and 8-bit per-token activation quantization, and (iii) low-bit weight-only quantization schemes. Our code is available at Software.

Jung Hyun Lee, June Yong Yang, Jungwook Choi et al.

As large language models continue to scale, low-bit weight-only post-training quantization (PTQ) offers a practical solution to their memory-efficient deployment. Although block-wise PTQ is capable of matching the full-precision (FP) baseline on basic language modeling and understanding, its quality is degraded for generative tasks -- especially at longer responses and extended chains of thought, which is critical in boosting task accuracy. We attribute this shortfall to two factors: (i) the omission of the unembedding layer (the LM head) in block-wise optimization and (ii) the reliance on the mean squared error (MSE) objective. Both factors cause the token probability distribution of the quantized model to misalign with that of the FP model, yielding notable accuracy drops on text generation benchmarks. To rectify the discrepancy, we introduce Logit-aware Final-block Quantization (LFQ), a simple yet effective enhancement to block-wise PTQ that quantizes the final Transformer block by minimizing the cross-entropy between the logits of the FP model and those of its quantized counterpart. By aligning token probabilities at the logit level in the final block, LFQ consistently improves the accuracy of complex generation tasks over state-of-the-art block-wise PTQ across diverse model families, while maintaining parity with FP baselines on language modeling and understanding.

Jung Hyun Lee, June Yong Yang, Byeongho Heo et al.

With the rapid advancement of test-time compute search strategies to improve the mathematical problem-solving capabilities of large language models (LLMs), the need for building robust verifiers has become increasingly important. However, all these inference strategies rely on existing verifiers originally designed for Best-of-N search, which makes them sub-optimal for tree search techniques at test time. During tree search, existing verifiers can only offer indirect and implicit assessments of partial solutions or under-value prospective intermediate steps, thus resulting in the premature pruning of promising intermediate steps. To overcome these limitations, we propose token-supervised value models (TVMs) - a new class of verifiers that assign each token a probability that reflects the likelihood of reaching the correct final answer. This new token-level supervision enables TVMs to directly and explicitly evaluate partial solutions, effectively distinguishing between promising and incorrect intermediate steps during tree search at test time. Experimental results demonstrate that combining tree-search-based inference strategies with TVMs significantly improves the accuracy of LLMs in mathematical problem-solving tasks, surpassing the performance of existing verifiers.

Changhun Kim, Taewon Kim, Seungyeon Woo et al.

In real-world scenarios, tabular data often suffer from distribution shifts that threaten the performance of machine learning models. Despite its prevalence and importance, handling distribution shifts in the tabular domain remains underexplored due to the inherent challenges within the tabular data itself. In this sense, test-time adaptation (TTA) offers a promising solution by adapting models to target data without accessing source data, crucial for privacy-sensitive tabular domains. However, existing TTA methods either 1) overlook the nature of tabular distribution shifts, often involving label distribution shifts, or 2) impose architectural constraints on the model, leading to a lack of applicability. To this end, we propose AdapTable, a novel TTA framework for tabular data. AdapTable operates in two stages: 1) calibrating model predictions using a shift-aware uncertainty calibrator, and 2) adjusting these predictions to match the target label distribution with a label distribution handler. We validate the effectiveness of AdapTable through theoretical analysis and extensive experiments on various distribution shift scenarios. Our results demonstrate AdapTable's ability to handle various real-world distribution shifts, achieving up to a 16% improvement on the HELOC dataset.

Yoonjeon Kim, Soohyun Ryu, Yeonsung Jung et al.

The development of vision-language and generative models has significantly advanced text-guided image editing, which seeks the preservation of core elements in the source image while implementing modifications based on the target text. However, existing metrics have a context-blindness problem, indiscriminately applying the same evaluation criteria on completely different pairs of source image and target text, biasing towards either modification or preservation. Directional CLIP similarity, the only metric that considers both source image and target text, is also biased towards modification aspects and attends to irrelevant editing regions of the image. We propose AugCLIP, a context-aware metric that adaptively coordinates preservation and modification aspects, depending on the specific context of a given source image and target text. This is done by deriving the CLIP representation of an ideally edited image, that preserves the source image with necessary modifications to align with target text. More specifically, using a multi-modal large language model, AugCLIP augments the textual descriptions of the source and target, then calculates a modification vector through a hyperplane that separates source and target attributes in CLIP space. Extensive experiments on five benchmark datasets, encompassing a diverse range of editing scenarios, show that AugCLIP aligns remarkably well with human evaluation standards, outperforming existing metrics. The code is available at https://github.com/augclip/augclip_eval.

Hyunseung Lim, Sooyohn Nam, Sungmin Na et al.

Patent examination remains an ongoing challenge in the NLP literature even after the advent of large language models (LLMs), as it requires an extensive yet nuanced human judgment on whether a submitted claim meets the statutory standards of novelty and non-obviousness against previously granted claims -- prior art -- in expert domains. Previous NLP studies have approached this challenge as a prediction task (e.g., forecasting grant outcomes) with high-level proxies such as similarity metrics or classifiers trained on historical labels. However, this approach often overlooks the step-by-step evaluations that examiners must make with profound information, including rationales for the decisions provided in office actions documents, which also makes it harder to measure the current state of techniques in patent review processes. To fill this gap, we construct PANORAMA, a dataset of 8,143 U.S. patent examination records that preserves the full decision trails, including original applications, all cited references, Non-Final Rejections, and Notices of Allowance. Also, PANORAMA decomposes the trails into sequential benchmarks that emulate patent professionals' patent review processes and allow researchers to examine large language models' capabilities at each step of them. Our findings indicate that, although LLMs are relatively effective at retrieving relevant prior art and pinpointing the pertinent paragraphs, they struggle to assess the novelty and non-obviousness of patent claims. We discuss these results and argue that advancing NLP, including LLMs, in the patent domain requires a deeper understanding of real-world patent examination. Our dataset is openly available at https://huggingface.co/datasets/LG-AI-Research/PANORAMA.

June Yong Yang, Byeongwook Kim, Jeongin Bae et al.

Key-Value (KV) Caching has become an essential technique for accelerating the inference speed and throughput of generative Large Language Models~(LLMs). However, the memory footprint of the KV cache poses a critical bottleneck in LLM deployment as the cache size grows with batch size and sequence length, often surpassing even the size of the model itself. Although recent methods were proposed to select and evict unimportant KV pairs from the cache to reduce memory consumption, the potential ramifications of eviction on the generative process are yet to be thoroughly examined. In this paper, we examine the detrimental impact of cache eviction and observe that unforeseen risks arise as the information contained in the KV pairs is exhaustively discarded, resulting in safety breaches, hallucinations, and context loss. Surprisingly, we find that preserving even a small amount of information contained in the evicted KV pairs via reduced precision quantization substantially recovers the incurred degradation. On the other hand, we observe that the important KV pairs must be kept at a relatively higher precision to safeguard the generation quality. Motivated by these observations, we propose \textit{Mixed-precision KV cache}~(MiKV), a reliable cache compression method that simultaneously preserves the context details by retaining the evicted KV pairs in low-precision and ensure generation quality by keeping the important KV pairs in high-precision. Experiments on diverse benchmarks and LLM backbones show that our proposed method offers a state-of-the-art trade-off between compression ratio and performance, compared to other baselines.

Yeonsung Jung, Jaeyun Song, June Yong Yang et al.

Learning generalized models from biased data is an important undertaking toward fairness in deep learning. To address this issue, recent studies attempt to identify and leverage bias-conflicting samples free from spurious correlations without prior knowledge of bias or an unbiased set. However, spurious correlation remains an ongoing challenge, primarily due to the difficulty in precisely detecting these samples. In this paper, inspired by the similarities between mislabeled samples and bias-conflicting samples, we approach this challenge from a novel perspective of mislabeled sample detection. Specifically, we delve into Influence Function, one of the standard methods for mislabeled sample detection, for identifying bias-conflicting samples and propose a simple yet effective remedy for biased models by leveraging them. Through comprehensive analysis and experiments on diverse datasets, we demonstrate that our new perspective can boost the precision of detection and rectify biased models effectively. Furthermore, our approach is complementary to existing methods, showing performance improvement even when applied to models that have already undergone recent debiasing techniques.

Joonhyung Park, June Yong Yang, Jinwoo Shin et al.

The Mixup scheme suggests mixing a pair of samples to create an augmented training sample and has gained considerable attention recently for improving the generalizability of neural networks. A straightforward and widely used extension of Mixup is to combine with regional dropout-like methods: removing random patches from a sample and replacing it with the features from another sample. Albeit their simplicity and effectiveness, these methods are prone to create harmful samples due to their randomness. To address this issue, 'maximum saliency' strategies were recently proposed: they select only the most informative features to prevent such a phenomenon. However, they now suffer from lack of sample diversification as they always deterministically select regions with maximum saliency, injecting bias into the augmented data. In this paper, we present, a novel, yet simple Mixup-variant that captures the best of both worlds. Our idea is two-fold. By stochastically sampling the features and 'grafting' them onto another sample, our method effectively generates diverse yet meaningful samples. Its second ingredient is to produce the label of the grafted sample by mixing the labels in a saliency-calibrated fashion, which rectifies supervision misguidance introduced by the random sampling procedure. Our experiments under CIFAR, Tiny-ImageNet, and ImageNet datasets show that our scheme outperforms the current state-of-the-art augmentation strategies not only in terms of classification accuracy, but is also superior in coping under stress conditions such as data corruption and object occlusion.

Yeonsung Jung, Hajin Shim, June Yong Yang et al.

Deep neural networks (DNNs), despite their impressive ability to generalize over-capacity networks, often rely heavily on malignant bias as shortcuts instead of task-related information for discriminative tasks. To address this problem, recent studies utilize auxiliary information related to the bias, which is rarely obtainable in practice, or sift through a handful of bias-free samples for debiasing. However, the success of these methods is not always guaranteed due to the unfulfilled presumptions. In this paper, we propose a novel method, Contrastive Debiasing via Generative Bias-transformation (CDvG), which works without explicit bias labels or bias-free samples. Motivated by our observation that not only discriminative models but also image translation models tend to focus on the malignant bias, CDvG employs an image translation model to transform one bias mode into another while preserving the task-relevant information. Additionally, the bias-transformed views are set against each other through contrastive learning to learn bias-invariant representations. Our method demonstrates superior performance compared to prior approaches, especially when bias-free samples are scarce or absent. Furthermore, CDvG can be integrated with the methods that focus on bias-free samples in a plug-and-play manner for additional enhancements, as demonstrated by diverse experimental results.

Geondo Park, June Yong Yang, Sung Ju Hwang et al.

Neural networks embedded in safety-sensitive applications such as self-driving cars and wearable health monitors rely on two important techniques: input attribution for hindsight analysis and network compression to reduce its size for edge-computing. In this paper, we show that these seemingly unrelated techniques conflict with each other as network compression deforms the produced attributions, which could lead to dire consequences for mission-critical applications. This phenomenon arises due to the fact that conventional network compression methods only preserve the predictions of the network while ignoring the quality of the attributions. To combat the attribution inconsistency problem, we present a framework that can preserve the attributions while compressing a network. By employing the Weighted Collapsed Attribution Matching regularizer, we match the attribution maps of the network being compressed to its pre-compression former self. We demonstrate the effectiveness of our algorithm both quantitatively and qualitatively on diverse compression methods.