This paper describes our system submitted to SemEval-2025 Task 4, which introduces the Synthetic Token Alternative Training (STAT) algorithm for efficient unlearning in large language models (LLMs). The proposed method aims to enable pretrained models to selectively forget designated data (the forget set) while preserving performance on the remaining data (the retain set).The STAT framework adopts a dual-stage process. In the first stage, pseudo tokens are generated through random sampling and applied to the forget set, facilitating more effective targeted unlearning. In the second stage, the model undergoes gradient-based optimization using an alternative training scheme that alternates between pseudo-token-augmented samples from the forget set and unmodified samples from the retain set. This design promotes stable unlearning of the specified data while accelerating convergence and preserving the model’s general performance.Our system achieved 3rd place in the 7B model track (OLMo-7B) and 7th place in the 1B model track (OLMo-1B), demonstrating substantial improvements over the official baselines, exhibiting superior stability in knowledge retention and more effective targeted forgetting compared to existing approaches.

This paper presents a hierarchical classification framework, designated as the Tree-guided Stagewise Classifier (TGSC) , which implements a Chain-of-Thought (CoT) reasoning paradigm for addressing multi-label and multi-class classification challenges in multilingual news article analysis in SemEval-2025 Task 10. The proposed methodology leverages the zero-shot capabilities inherent in Large Language Models (LLMs) through a systematic hierarchical reasoning mechanism. This process proceeds through successive hierarchical levels, wherein the classification commences from root nodes and progressively navigates category branches via iterative determinations at each hierarchical tier, ultimately culminating in leaf category identification during the final classification stage. To optimize classification precision, a specialized prompt engineering strategy incorporating hierarchical structural constraints is developed to guide the reasoning trajectory. Experimental results demonstrate the effectiveness of our approach, achieving competitive performance across multiple languages in Subtask 1 and Subtask 2.

Zero-shot dialogue state tracking (DST) seeks to enable dialogue systems to transition to unfamiliar domains without manual annotation or extensive retraining. Prior research has approached this objective by embedding prompts into language models (LMs). Common methodologies include integrating prompts at the input layer or introducing learnable variables at each transformer layer. Nonetheless, each strategy exhibits inherent limitations. Prompts integrated at the input layer risk underutilization, with their impact potentially diminishing across successive transformer layers. Conversely, the addition of learnable variables to each layer can complicate the training process and increase inference latency. To tackle the issues mentioned above, this paper proposes Dual Low-Rank Adaptation (DualLoRA), a plug-and-play architecture designed for zero-shot DST. DualLoRA incorporates two distinct Low-Rank Adaptation (LoRA) components, targeting both dialogue context processing and prompt optimization, to ensure the comprehensive influence of prompts throughout the transformer model layers. This is achieved without incurring additional inference latency, showcasing an efficient integration into existing architectures. Through rigorous evaluation on the MultiWOZ and SGD datasets, DualLoRA demonstrates notable improvements across multiple domains, outperforming traditional baseline methods in zero-shot settings.

User Simulators play a pivotal role in training and evaluating task-oriented dialogue systems. Traditional user simulators typically rely on human-engineered agendas, resulting in generated responses that often lack diversity and spontaneity. Although large language models (LLMs) exhibit a remarkable capacity for generating coherent and contextually appropriate utterances, they may fall short when tasked with generating responses that effectively guide users towards their goals, particularly in dialogues with intricate constraints and requirements. This paper introduces DuetSim, a novel framework designed to address the intricate demands of task-oriented dialogues by leveraging LLMs. DuetSim stands apart from conventional approaches by employing two LLMs in tandem: one dedicated to response generation and the other focused on verification. This dual LLM approach empowers DuetSim to produce responses that not only exhibit diversity but also demonstrate accuracy and are preferred by human users. We validate the efficacy of our method through extensive experiments conducted on the MultiWOZ dataset, highlighting improvements in response quality and correctness, largely attributed to the incorporation of the second LLM.