As Large Language Models (LLMs) scale up and gain powerful Chain-of-Thoughts (CoTs) reasoning abilities, practical resource constraints drive efforts to distill these capabilities into more compact Smaller Language Models (SLMs).We find that CoTs consist mainly of simple reasoning forms, with a small proportion (≈ 4.7%) of key reasoning steps that truly impact conclusions.However, previous distillation methods typically involve supervised fine-tuning student SLMs only on correct CoTs data produced by teacher LLMs, resulting in students struggling to learn the key, instead imitating the teacher's reasoning forms and making errors or omissions in reasoning.To address these issues, drawing an analogy to human learning, where analyzing mistakes according to correct solutions often reveals the crucial steps leading to successes or failures, we propose mistakE-Driven key reasonIng step distillaTion (EDIT), a novel method that further aids SLMs learning key reasoning steps rather than mere simple fine-tuning.Firstly, to expose the crucial steps in CoTs, we carefully design specific prompts to generate dual CoTs data with similar reasoning paths but divergent conclusions.Then, we apply the minimum edit distance algorithm on the dual CoTs data to locate these key steps and optimize the likelihood on these tokens.Extensive experiments and analysis validate the effectiveness of EDIT across both in-domain(IND) and out-ofdomain(OOD) benchmark reasoning datasets 1 .* Kun Li is the corresponding author. 1 Code can be found at https://github.com/ C-W-D/EDIT On the floor, I see two magenta scrunchiephone chargers, one gold fidget spinner, three gold booklets, three silver scrunchiephone chargers, two silver booklets, and two magenta booklets.If I remove all the magenta items from the floor, how many scrunchiephone chargers remain on it?Op ons: (A) zero (B) one (C) two (D) three (E) four (F) five (G) six According to this ques on, there are two magenta scrunchiephone chargers and three silver scrunchiephone chargers on the floor.If we remove all the magenta items from the floor, we are le with 2 silver scrunchiephone chargers.Therefore, the answer is (C).According to this ques on, there are two magenta scrunchiephone chargers and three silver scrunchiephone chargers on the floor.If we remove all the magenta items, we are le with only the silver scrunchiephone chargers.Therefore, the number of scrunchiephone chargers that remain on the floor is three.Therefore, the answer is (D).Teacher Ques on Unthinking Student On the floor, I see two magenta scrunchiephone chargers, one gold fidget spinner...According to this ques on, there are two magenta scrunchiephone chargers and three silver scrunchiephone chargers on the floor.If we remove all the magenta items from the floor, we are le with 2 silver scrunchiephone chargers.Therefore, the answer is (C).According to this ques on, there are a total of 5 scrunchiephone chargers on the floor: 2 magentascrunchiephone chargers and 3 silver scrunchiephone chargers.If we remove all the magenta items from the floor, we remove the 2 magenta scrunchiephone chargers.Therefore, the number of scrunchiephone chargers remaining on the floor is 3. Therefore, the answer is (D). Key Steps in Correct CoTs Data Key Steps in Wrong CoTs Data

Large language models (LLMs) exhibit enhanced reasoning at larger scales, driving efforts to distill these capabilities into smaller models via teacher-student learning.Previous works simply fine-tune student models on teachers' generated Chain-of-Thoughts (CoTs) data.Although these methods enhance indomain (IND) reasoning performance, they struggle to generalize to out-of-domain (OOD) tasks.We believe that the widespread spurious correlations between questions and answers may lead the model to preset a specific answer which restricts the diversity and generalizability of its reasoning process.In this paper, we propose Cascading Decomposed CoTs Distillation (CasCoD) to address these issues by decomposing the traditional single-step learning process into two cascaded learning steps.Specifically, by restructuring the training objectives-removing the answer from outputs and concatenating the question with the rationale as input-CasCoD's two-step learning process ensures that students focus on learning rationales without interference from the preset answers, thus improving reasoning generalizability.Extensive experiments demonstrate the effectiveness of CasCoD on both IND and OOD benchmark reasoning datasets 1 .* Kun Li is the corresponding author. 1 Code available at https://github.com/C-W-D/CasCoD(a) Answer SFT consistently outperform Std-CoT on OOD tasks.(b) A case of spurious correla on between ques ons and answers.Question: Why did someone bring a swimsuit to a ski resort?Options: (A) To swim in a heated pool.(B) To wear as an underlayer for warmth.(C) To use as a fashion statement.(D) To participate in a polar bear plunge event.Answer: (A) To swim in a heated pool. OOD IND

We study model extraction attacks in natural language processing (NLP) where attackers aim to steal victim models by repeatedly querying the open Application Programming Interfaces (APIs).Recent works focus on limited-query budget settings and adopt random sampling or active learning-based sampling strategies on publicly available, unannotated data sources.However, these methods often result in selected queries that lack task relevance and data diversity, leading to limited success in achieving satisfactory results with low query costs.In this paper, we propose MeaeQ (Model extraction attack with efficient Queries), a straightforward yet effective method to address these issues.Specifically, we initially utilize a zero-shot sequence inference classifier, combined with API service information, to filter task-relevant data from a public text corpus instead of a problem domain-specific dataset.Furthermore, we employ a clustering-based data reduction technique to obtain representative data as queries for the attack.Extensive experiments conducted on four benchmark datasets demonstrate that MeaeQ achieves higher functional similarity to the victim model than baselines while requiring fewer queries.Our code is available at https://github.com/C-W-D/MeaeQ.

Neural network models are vulnerable to adversarial examples, and adversarial transferability further increases the risk of adversarial attacks.Current methods based on transferability often rely on substitute models, which can be impractical and costly in real-world scenarios due to the unavailability of training data and the victim model's structural details.In this paper, we propose a novel approach that directly constructs adversarial examples by extracting transferable features across various tasks.Our key insight is that adversarial transferability can extend across different tasks.Specifically, we train a sequence-to-sequence generative model named CT-GAT (Cross-Task Generative Adversarial ATtack) using adversarial sample data collected from multiple tasks to acquire universal adversarial features and generate adversarial examples for different tasks.We conduct experiments on ten distinct datasets, and the results demonstrate that our method achieves superior attack performance with small cost.You can get our code and data at: https: //github.com/xiaoxuanNLP/CT-GAT