DaYou Du, Yijia Zhang, Shijie Cao, Jiaqi Guo, Ting Cao, Xiaowen Chu, Ningyi Xu. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Much attention is being paid to using high-performance convolutional neural networks (CNNs) in the area of ship detection in optical remoting sensing (ORS) images. However, the problem of false negatives (FNs) caused by side-by-side ships cannot be solved, and the number of false positives (FPs) remains high. This paper uses a DLA-34 network with deformable convolution layers as the backbone. The network has two priority branches: a recall-priority branch for reducing the number of FNs, and a precision-priority branch for reducing the number of FPs. In our single-shot detection method, the recall-priority branch is based on an anchor-free module without non-maximum suppression (NMS), while the precision-priority branch utilizes an anchor-based module with NMS. We perform recall-priority branch functions based on the output part of the CenterNet object detector to precisely predict center points of bounding boxes. The Bidirectional Feature Pyramid Network (BiFPN), combined with the inference part of YOLOv3, is used to improve the precision of precision-priority branch. Finally, the boxes from two branches merge, and we propose priority-based selection (PBS) for choosing the accurate ones. Results show that our proposed method sharply improves the recall rate of side-by-side ships and significantly reduces the number of false alarms. Our method also achieves the best trade-off on our improved version of HRSC2016 dataset, with 95.57% AP at 56 frames per second on an Nvidia RTX-2080 Ti GPU. Compared with the HRSC2016 dataset, not only are our annotations more accurate, but our dataset also contains more images and samples. Our evaluation metrics also included tests on small ships and incomplete forms of ships.

Yijia Zhang, Sicheng Zhang, Shijie Cao, DaYou Du, Jianyu Wei, Ting Cao, Ningyi Xu. Findings of the Association for Computational Linguistics ACL 2024. 2024.

Efficient deployment of large language models (LLMs) necessitates low-bit quantization to minimize model size and inference cost. While low-bit integer formats (e.g., INT8/INT4) have been the conventional choice, emerging low-bit floating-point formats (e.g., FP8/FP4) offer a compelling alternative and are gaining support from cutting-edge hardware, such as NVIDIA's H100 GPU. However, the superiority of low-bit INT versus FP formats for quantization on LLMs remains unclear. In this study, we conduct a comparative analysis of INT and FP quantization with the same bit-width, revealing that the optimal quantization format varies across different layers due to the complexity and diversity of tensor distribution. Consequently, we advocate the Mixture of Formats Quantization (MoFQ), which selects the optimal format on a layer-wise basis. This simple yet effective approach achieves state-of-the-art results in both weight-only (W-only) and weight-activation (WA) post-training quantization scenarios when tested on LLaMA across various tasks. In 4-bit W-only quantization, MoFQ surpasses GPTQ without complex hyperparameter tuning and with an order of magnitude faster quantization speed. While in 8-bit WA quantization, MoFQ significantly outperforms INT/FP-only methods, achieving performance close to the full precision model. Notably, MoFQ incurs no hardware overhead compared to INT/FP-only quantization, as the bit-width remains unchanged.

Large language models (LLMs) show great performance in various tasks, but face deployment challenges from limited memory capacity and bandwidth. Low-bit weight quantization can save memory and accelerate inference. Although floating-point (FP) formats show good performance in LLM quantization, they tend to perform poorly with small group sizes or sub-4 bits. We find the reason is that the absence of asymmetry in previous FP quantization makes it unsuitable for handling asymmetric value distribution of LLM weight tensors. In this work, we propose asymmetric FP quantization (AFPQ), which sets separate scales for positive and negative values. Our method leads to large accuracy improvements and can be easily plugged into other quantization methods, including GPTQ and AWQ, for better performance. Besides, no additional storage is needed compared with asymmetric integer (INT) quantization. The code is available at https://github.com/zhangsichengsjtu/AFPQ.

The upscaling of Large Language Models (LLMs) has yielded impressive advances in natural language processing, yet it also poses significant deployment challenges. Weight quantization has emerged as a widely embraced solution to reduce memory and computational demands. This paper introduces BitDistiller, a framework that synergizes Quantization-Aware Training (QAT) with Knowledge Distillation (KD) to boost the performance of LLMs at ultra-low precisions (sub-4-bit). Specifically, BitDistiller first incorporates a tailored asymmetric quantization and clipping technique to maximally preserve the fidelity of quantized weights, and then proposes a novel Confidence-Aware Kullback-Leibler Divergence (CAKLD) objective, which is employed in a self-distillation manner to enable faster convergence and superior model performance. Empirical evaluations demonstrate that BitDistiller significantly surpasses existing methods in both 3-bit and 2-bit configurations on general language understanding and complex reasoning benchmarks. Notably, BitDistiller is shown to be more cost-effective, demanding fewer data and training resources. The code is available at https://github.com/DD-DuDa/BitDistiller.