Vertical-domain few-shot classification remains challenging for small language models, as limited supervision makes it difficult to acquire domain-specific decision knowledge. On-Policy Distillation (OPD) can improve teacher-guided adaptation by supervising student-generated rollouts, while GRPO-based reinforcement learning can further refine downstream predictions. However, existing KD-to-RL pipelines typically rely on globally fixed transition schedules, ignoring that different samples may require different amounts of teacher-guided acquisition before reward-driven refinement. We propose PIVOT (Perplexity-Informed Transition Optimization), a dynamic transition framework that routes samples between OPD and GRPO according to teacher-evaluated sequence perplexity. PIVOT moves low-perplexity samples to GRPO for reward-driven refinement while keeping high-perplexity samples under OPD for continued domain knowledge acquisition. Experiments on Banking77 and HWU64 show that PIVOT consistently outperforms continued OPD and globally synchronized OPD→GRPO baselines under the same optimization budget, achieving stronger downstream performance and more stable training dynamics.

Parameter-Efficient Fine-Tuning (PEFT) methods, especially LoRA, are widely used for adapting pre-trained models to downstream tasks due to their computational and storage efficiency. However, in the context of LoRA and its variants, the potential of activation subspaces corresponding to tail eigenvectors remains substantially under-exploited, which may lead to suboptimal fine-tuning performance. In this work, we propose Astra (Activation-Space Tail-Eigenvector Low-Rank Adaptation), a novel PEFT method that leverages the tail eigenvectors of the model output activations—estimated from a small task-specific calibration set—to construct task-adaptive low-rank adapters. By constraining updates to the subspace spanned by these tail eigenvectors, Astra achieves faster convergence and improved downstream performance with a significantly reduced parameter budget. Extensive experiments across natural language understanding (NLU) and natural language generation (NLG) tasks demonstrate that Astra consistently outperforms existing PEFT baselines across 16 benchmarks and even surpasses full fine-tuning (FFT) in certain scenarios.

Voice agents typically use text-first pipeline for safety, but large language model (LLM) outputs are often overly written, making even high-quality text-to-speech (TTS) sound robotic. This is especially problematic in domains such as healthcare and sales, where different audiences require graded orality control. Text-level rewriting can control spoken style, but resources mostly study the reverse direction (spoken-to-written transfer), and its internal locus in LLMs remains unclear. We study English written-to-spoken rewriting with graded orality control. We introduce WSPairs, 4,709 written–spoken pairs from spontaneous conversations, annotated with 0–5 Orality Levels using seven sociolinguistic features. Through causal activation patching and head-level attribution, we find that orality concentrates in late layers and a sparse set of high-influence attention heads. We then propose CAS, a no-weight-update calibrated activation-steering method, and LAS, a learned activation-steering method. CAS improves exact level targeting and reduces bias relative to few-shot prompting, although target ordering remains imperfect at the highest levels. LAS uses 0.004%–0.006% trainable parameters, obtains higher Orality Level Accuracy (L-Acc) than full SFT on Qwen-3B and Mistral-7B, remains the strongest lightweight method on Llama-1B, and leads SFT by up to 3.9 L-Acc points out of domain. These results support localized, efficient control of text-level spoken style. Dataset and code are available.

Currently, most agentic systems are driven by mainstream closed-source large language models (LLMs), such as GPT-5 and Gemini 3, resulting in high operational costs. As a result, deploying customized small language models (SLMs) to replace LLMs in agent components has attracted increasing attention. However, the limited capabilities of SLMs often render them inadequate for agent tasks. To address this, we propose Edit4Agency, a novel framework that utilizes model editing to efficiently and effectively enhance agent sub-capabilities of SLMs. By formalizing tasks as a sequence generation problem triggered by specific conditions, the framework embeds task triggers into system prompts and performs model editing through localized parameter updates, eliminating the need for large-scale retraining. Extensive experiments across multiple datasets and case studies in agent systems demonstrate that model editing, with only 50 data samples, enables a base model without instruction tuning to acquire new skills such as tool calling while maintaining its general capabilities. The results show that Edit4Agency outperforms instruction-tuned SLMs, constrained decoding, and SLMs trained with large-scale supervised fine-tuning or reinforcement learning methods.

In recent years, large language models (LLMs) have made significant progress in knowledge-intensive applications. However, when adapting them to specific domains, we may encounter a multi-stage continuous learning scenario, especially in cases where domain knowledge evolves rapidly.This issue severely limits traditional fine-tuning approaches for LLMs.To overcome this limitation, we propose a new learning paradigm designed specifically for multi-stage continuous learning. This paradigm includes a preference-based learning bias to identify potential knowledge conflicts, as well as a self-distillation-based data augmentation strategy to expand and enrich the training corpus, thereby improving the integration of knowledge-compatible information.In the experiments, we show that our proposed method achieves a significant improvement in accuracy after 7 stages of fine-tuning compared to previous methods, while also demonstrating excellent performance in preserving general knowledge.We have released our code and dataset at Multi-Stage-Learning.

Large language models (LLMs) often exhibit Context Faithfulness Hallucinations, where outputs deviate from retrieved information due to incomplete context integration. Our analysis reveals a strong correlation between token-level uncertainty and hallucinations. We hypothesize that attention mechanisms inherently encode context utilization signals, supported by probing analysis. Based on these insights, we propose Dynamic Attention-Guided Context Decoding (DAGCD), a lightweight framework that leverages attention distributions and uncertainty signals in a single-pass decoding. Experiments on open-book QA datasets demonstrate DAGCD’s effectiveness, yielding significant improvements in faithfulness and robustness while preserving computational efficiency.

Recent studies have demonstrated that many layers are functionally redundant in large language models (LLMs), enabling model compression by removing these layers to reduce inference cost. While such approaches can improve efficiency, indiscriminate layer pruning often results in significant performance degradation. In this paper, we propose GRASP (Gradient-based Retention of Adaptive Singular Parameters), a novel compression framework that mitigates this issue by preserving sensitivity-aware singular values. Unlike direct layer pruning, GRASP leverages gradient-based attribution on a small calibration dataset to adaptively identify and retain critical singular components. By replacing redundant layers with only a minimal set of parameters, GRASP achieves efficient compression while maintaining strong performance with minimal overhead. Experiments across multiple LLMs show that GRASP consistently outperforms existing compression methods, achieving 90% of the original model’s performance under a 20% compression ratio.

In the realm of Large Language Models (LLMs), the balance between instruction data quality and quantity is a focal point. Recognizing this, we introduce a self-guided methodology for LLMs to autonomously discern and select cherry samples from open-source datasets, effectively minimizing manual curation and potential cost for instruction tuning an LLM. Our key innovation, the Instruction-Following Difficulty (IFD) metric, emerges as a pivotal metric to identify discrepancies between a model’s expected responses and its intrinsic generation capability. Through the application of IFD, cherry samples can be pinpointed, leading to a marked uptick in model training efficiency. Empirical validations on datasets like Alpaca and WizardLM underpin our findings; with a mere 10% of original data input, our strategy showcases improved results. This synthesis of self-guided cherry-picking and the IFD metric signifies a transformative leap in the instruction tuning of LLMs, promising both efficiency and resource-conscious advancements. Codes, data, and models are available.

In order to construct or extend entity-centric and event-centric knowledge graphs (KG and EKG), the information extraction (IE) annotation toolkit is essential. However, existing IE toolkits have several non-trivial problems, such as not supporting multi-tasks, and not supporting automatic updates. In this work, we present CollabKG, a learnable human-machine-cooperative IE toolkit for KG and EKG construction. Specifically, for the multi-task issue, CollabKG unifies different IE subtasks, including named entity recognition (NER), entity-relation triple extraction (RE), and event extraction (EE), and supports both KG and EKG. Then, combining advanced prompting-based IE technology, the human-machine-cooperation mechanism with Large Language Models (LLMs) as the assistant machine is presented which can provide a lower cost as well as a higher performance. Lastly, owing to the two-way interaction between the human and machine, CollabKG with learning ability allows self-renewal. Besides, CollabKG has several appealing features (e.g., customization, training-free, and label propagation) that make the system powerful and high-productivity. We holistically compare our toolkit with other existing tools on these features. Human evaluation quantitatively illustrates that CollabKG significantly improves annotation quality, efficiency, and stability simultaneously.

“Deploying tuning-free prompting is challenging in engineering practice: it not only requiresusers to engage in cumbersome trials and errors but is also extremely time-consuming,as even a slight change in wording and phrasing could have a huge impact on the finalperformance. To further investigate the impact of different prompts, in this work, weperform a systematic inspection of four factors in linguistics involved in prompt engineering:syntax, semantics, lexicon, and pragmatics. The empirical results quantify the sensitivityof the output to small textual perturbations in four linguistic factors of prompts. Basedon the analysis of these four factors, we present a series of design guidelines to helphuman users write effective prompts. Human evaluation on amateurs shows that usingthe proposed guidelines helps humans produce prompts with significant gains in zero-shotperformance in Pre-trained Language Models (PLMs) and hence validates the utility ofthe guidelines.”

Instruction tuning is critical to improve LLMs but usually suffers from low-quality and redundant data. Data filtering for instruction tuning has proved important in improving both the efficiency and performance of the tuning process. But it also leads to extra cost and computation due to the involvement of LLMs in this process. To reduce the filtering cost, we study Superfiltering: Can we use a smaller and weaker model to select data for finetuning a larger and stronger model? Despite the performance gap between weak and strong language models, we find their highly consistent capability to perceive instruction difficulty and data selection results. This enables us to use a much smaller and more efficient model to filter the instruction data used to train a larger language model. Not only does it largely speed up the data filtering, but the filtered-data-finetuned LLM achieves even better performance on standard benchmarks. Extensive experiments validate the efficacy and efficiency of our approach.

The Retrieval Question Answering (ReQA) task employs the retrieval-augmented framework, composed of a retriever and generator. The generators formulate the answer based on the documents retrieved by the retriever. Incorporating Large Language Models (LLMs) as generators is beneficial due to their advanced QA capabilities, but they are typically too large to be fine-tuned with budget constraints while some of them are only accessible via APIs. To tackle this issue and further improve ReQA performance, we propose a trainable Pluggable Reward-Driven Contextual Adapter (PRCA), keeping the generator as a black box. Positioned between the retriever and generator in a Pluggable manner, PRCA refines the retrieved information by operating in a token-autoregressive strategy via maximizing rewards of the reinforcement learning phase. Our experiments validate PRCA’s effectiveness in enhancing ReQA performance on three datasets by up to 20% improvement to fit black-box LLMs into existing frameworks, demonstrating its considerable potential in the LLMs era.

The basic tasks of ancient Chinese information processing include automatic sentence segmentation, word segmentation, part-of-speech tagging and named entity recognition. Tasks such as lexical analysis need to be based on sentence segmentation because of the reason that a plenty of ancient books are not punctuated. However, step-by-step processing is prone to cause multi-level diffusion of errors. This paper designs and implements an integrated annotation system of sentence segmentation and lexical analysis. The BiLSTM-CRF neural network model is used to verify the generalization ability and the effect of sentence segmentation and lexical analysis on different label levels on four cross-age test sets. Research shows that the integration method adopted in ancient Chinese improves the F1-score of sentence segmentation, word segmentation and part of speech tagging. Based on the experimental results of each test set, the F1-score of sentence segmentation reached 78.95, with an average increase of 3.5%; the F1-score of word segmentation reached 85.73%, with an average increase of 0.18%; and the F1-score of part-of-speech tagging reached 72.65, with an average increase of 0.35%.