Improving the efficiency of inference in Large Language Models (LLMs) is a critical area of research. Post-training Quantization (PTQ) is a popular technique, but it often faces challenges at low-bit levels, particularly in downstream tasks. Quantization-aware Training (QAT) can alleviate this problem, but it requires significantly more computational resources. To tackle this, we introduced Weight-Decomposed Low-Rank Quantization-Aware Training (DL-QAT), which merges the advantages of QAT while training only less than 1% of the total parameters. Specifically, we introduce a group-specific quantization magnitude to adjust the overall scale of each quantization group. Within each quantization group, we use LoRA matrices to update the weight size and direction in the quantization space. We validated the effectiveness of our method on the LLaMA and LLaMA2 model families. The results show significant improvements over our baseline method across different quantization granularities. For instance, for LLaMA-7B, our approach outperforms the previous state-of-the-art method by 4.2% in MMLU on 3-bit LLaMA-7B. Additionally, our quantization results on pre-trained models also surpass previous QAT methods, demonstrating the superior performance and efficiency of our approach.

Large language models (LLMs) are increasingly being adopted in a wide range of real-world applications. Despite their impressive performance, recent studies have shown that LLMs are vulnerable to deliberately crafted adversarial prompts even when aligned via Reinforcement Learning from Human Feedback or supervised fine-tuning. While existing defense methods focus on either detecting harmful prompts or reducing the likelihood of harmful responses through various means, defending LLMs against jailbreak attacks based on the inner mechanisms of LLMs remains largely unexplored. In this work, we investigate how LLMs respond to harmful prompts and propose a novel defense method termed Layer-specific Editing (LED) to enhance the resilience of LLMs against jailbreak attacks. Through LED, we reveal that several critical safety layers exist among the early layers of LLMs. We then show that realigning these safety layers (and some selected additional layers) with the decoded safe response from identified toxic layers can significantly improve the alignment of LLMs against jailbreak attacks. Extensive experiments across various LLMs (e.g., Llama2, Mistral) show the effectiveness of LED, which effectively defends against jailbreak attacks while maintaining performance on benign prompts. Our code is available at https://github.com/ledllm/ledllm.

The widespread use of large language models (LLMs) has sparked concerns about the potential misuse of AI-generated text, as these models can produce content that closely resembles human-generated text. Current detectors for AI-generated text (AIGT) lack robustness against adversarial perturbations, with even minor changes in characters or words causing a reversal in distinguishing between human-created and AI-generated text. This paper investigates the robustness of existing AIGT detection methods and introduces a novel detector, the Siamese Calibrated Reconstruction Network (SCRN). The SCRN employs a reconstruction network to add and remove noise from text, extracting a semantic representation that is robust to local perturbations. We also propose a siamese calibration technique to train the model to make equally confident predictions under different noise, which improves the model’s robustness against adversarial perturbations. Experiments on four publicly available datasets show that the SCRN outperforms all baseline methods, achieving 6.5%-18.25% absolute accuracy improvement over the best baseline method under adversarial attacks. Moreover, it exhibits superior generalizability in cross-domain, cross-genre, and mixed-source scenarios. The code is available at https://github.com/CarlanLark/Robust-AIGC-Detector.

Joint relational triple extraction from unstructured text is an important task in information extraction. However, most existing works either ignore the semantic information of relations or predict subjects and objects sequentially. To address the issues, we introduce a new blank filling paradigm for the task, and propose a relation-first blank filling network (RFBFN). Specifically, we first detect potential relations maintained in the text to aid the following entity pair extraction. Then, we transform relations into relation templates with blanks which contain the fine-grained semantic representation of the relations. Finally, corresponding subjects and objects are extracted simultaneously by filling the blanks. We evaluate the proposed model on public benchmark datasets. Experimental results show our model outperforms current state-of-the-art methods. The source code of our work is available at: https://github.com/lizhe2016/RFBFN.

Text classification tends to be difficult when data are inadequate considering the amount of manually labeled text corpora. For low-resource agglutinative languages including Uyghur, Kazakh, and Kyrgyz (UKK languages), in which words are manufactured via stems concatenated with several suffixes and stems are used as the representation of text content, this feature allows infinite derivatives vocabulary that leads to high uncertainty of writing forms and huge redundant features. There are major challenges of low-resource agglutinative text classification the lack of labeled data in a target domain and morphologic diversity of derivations in language structures. It is an effective solution which fine-tuning a pre-trained language model to provide meaningful and favorable-to-use feature extractors for downstream text classification tasks. To this end, we propose a low-resource agglutinative language model fine-tuning AgglutiFiT, specifically, we build a low-noise fine-tuning dataset by morphological analysis and stem extraction, then fine-tune the cross-lingual pre-training model on this dataset. Moreover, we propose an attention-based fine-tuning strategy that better selects relevant semantic and syntactic information from the pre-trained language model and uses those features on downstream text classification tasks. We evaluate our methods on nine Uyghur, Kazakh, and Kyrgyz classification datasets, where they have significantly better performance compared with several strong baselines.