Abstract The rapid evolution of large language models (LLMs) has driven a transformative shift in artificial intelligence (AI), reshaping both research paradigms and practical applications. Distinguished from their predecessors by unprecedented scale and advanced capabilities, LLMs necessitate new frameworks for understanding their development, behavior, and societal impact. This survey systematically reviews recent advancements in LLM techniques across four key dimensions: (1) pre-training methodologies, which establish core model capabilities through large-scale self-supervised training, architectural innovations, and data curation strategies; (2) post-training techniques, including supervised fine-tuning and reinforcement learning, which adapt foundational models to downstream tasks and enhance their alignment and safety; (3) utilization strategies, such as in-context learning, prompt engineering, and agentic reasoning, that optimize real-world deployment and enable effective interaction with external environments; and (4) evaluation methods, encompassing benchmarks for key ability dimensions such as core language capabilities, reasoning, and safety, which support comprehensive and reliable assessment of model performance. Additionally, we identify critical research issues, including those concerning theoretical foundations, efficient scaling, alignment, and agentic capability, and highlight the open challenges they present. By synthesizing state-of-the-art insights and emerging trends, this survey aims to provide a systematic and comprehensive framework for understanding the trajectory, current limitations, and future directions of LLM progress.

Zican Dong, Junyi Li, Jinhao Jiang, Mingyu Xu, Xin Zhao, Bingning Wang, Weipeng Chen. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Due to the immense resource demands and the involved complex techniques, it is still challenging for successfully pre-training a large language models (LLMs) with state-of-the-art performance. In this paper, we explore the key bottlenecks and designs during pre-training, and make the following contributions: (1) a comprehensive investigation into the factors contributing to training instability; (2) a robust optimization approach designed to mitigate training instability effectively; (3) an elaborate data pipeline that integrates data synthesis, data curriculum, and data selection. By integrating the above techniques, we create a rather low-cost training recipe and use it to pre-train YuLan-Mini, a fully-open base model with 2.4B parameters on 1.08T tokens. Remarkably, YuLan-Mini achieves top-tier performance among models of similar parameter scale, with comparable performance to industry-leading models that require significantly more data. To facilitate reproduction, we release the full details of training recipe and data composition. Project details can be accessed at the following link: https://anonymous.4open.science/r/YuLan-Mini/README.md.

Large reasoning models (LRMs) have exhibited strong performance on complex reasoning tasks, with further gains achievable through increased computational budgets at inference.However, current test-time scaling methods predominantly rely on redundant sampling, ignoring the historical experience utilization, thereby limiting computational efficiency.To overcome this limitation, we propose Sticker-TTS, a novel test-time scaling framework that coordinates three collaborative LRMs to iteratively explore and refine solutions guided by historical attempts.At the core of our framework are distilled key conditions-termed stickers-which drive the extraction, refinement, and reuse of critical information across multiple rounds of reasoning.To further enhance the efficiency and performance of our framework, we introduce a two-stage optimization strategy that combines imitation learning with self-improvement, enabling progressive refinement.Extensive evaluations on three challenging mathematical reasoning benchmarks, including AIME-24, AIME-25, and Olym-MATH, demonstrate that Sticker-TTS consistently surpasses strong baselines, including self-consistency and advanced reinforcement learning approaches, under comparable inference budgets.These results highlight the effectiveness of sticker-guided historical experience utilization.

Advancements in Large Language Models (LLMs) have extended their input context length, yet they still struggle with retrieval and reasoning in long-context inputs.Existing methods propose to utilize the prompt strategy and Retrieval-Augmented Generation (RAG) to alleviate this limitation.However, they still face challenges in balancing retrieval precision and recall, impacting their efficacy in answering questions.To address this, we introduce CAFE, a two-stage coarse-to-fine method to enhance multi-document questionanswering capacities.By gradually eliminating the negative impacts of background and distracting documents, CAFE makes the responses more reliant on the evidence documents.Initially, a coarse-grained filtering method leverages retrieval heads to identify and rank relevant documents.Then, a fine-grained steering method guides attention to the most relevant content.Experiments across benchmarks show that CAFE outperforms baselines, achieving an average SubEM improvement of up to 22.1% and 13.7% over SFT and RAG methods, respectively, across three different models.Our code is available at https://github. com/RUCAIBox/CAFE.

Mixture-of-Experts (MoE) models achieve a favorable trade-off between performance and inference efficiency by activating only a subset of experts. However, the memory overhead of storing all experts remains a major limitation, especially in large-scale MoE models such as DeepSeek-R1 (671B). In this study, we investigate domain specialization and expert redundancy in large-scale MoE models and uncover a consistent behavior we term~\emph{few-shot expert localization}, with only a few in-domain demonstrations, the model consistently activates a sparse and stable subset of experts on tasks within the same domain. Building on this observation, we propose a simple yet effective pruning framework, \textbf{EASY-EP}, that leverages a few domain-specific demonstrations to identify and retain only the most relevant experts. EASY-EP comprises two key components: \textbf{output-aware expert importance assessment} and \textbf{expert-level token contribution estimation}. The former evaluates the importance of each expert for the current token by considering the gating scores and L2 norm of the outputs of activated experts, while the latter assesses the contribution of tokens based on representation similarities before and after routed experts. Experiments on DeepSeek-R1 and DeepSeek-V3-0324 show that our method can achieve comparable performances and $2.99\times$ throughput under the same memory budget as the full model, with only half the experts. Our code is available at https://github.com/RUCAIBox/EASYEP.

Large language models (LLMs) have achieved dramatic proficiency over NLP tasks with normal length. Recently, multiple studies have committed to extending the context length and enhancing the long text modeling capabilities of LLMs. To comprehensively evaluate the long context ability of LLMs, we propose BAMBOO, a multi-task long context benchmark. BAMBOO has been designed with four principles: comprehensive capacity evaluation, avoidance of data contamination, accurate automatic evaluation, and different length levels. It consists of 10 datasets from 5 different long text understanding tasks, i.e., question answering, hallucination detection, text sorting, language modeling, and code completion, to cover various domains and core capacities of LLMs. We conduct experiments with five widely-used long-context models and further discuss five key questions for long text research. In the end, we discuss problems of current long-context models and point out future directions for enhancing long text modeling capacities. We release our data, prompts, and code at https://anonymous.4open.science/r/BAMBOO/.

Transformer-based large language models (LLMs) typically have a limited context window, resulting in significant performance degradation when processing text beyond the length of the context window. Extensive studies have been proposed to extend the context window and achieve length extrapolation of LLMs, but there is still a lack of in-depth interpretation of these approaches. In this study, we explore the positional information within and beyond the context window for deciphering the underlying mechanism of LLMs. By using a mean-based decomposition method, we disentangle positional vectors from hidden states of LLMs and analyze their formation and effect on attention. Furthermore, when texts exceed the context window, we analyze the change of positional vectors in two settings, i.e., direct extrapolation and context window extension. Based on our findings, we design two training-free context window extension methods, positional vector replacement and attention window extension. Experimental results show that our methods can effectively extend the context window length.

In this paper, we aim to improve the reasoning ability of large language models (LLMs) over structured data in a unified way.Inspired by the studies on tool augmentation for LLMs, we develop an Iterative Reading-then-Reasoning (IRR) framework to solve question answering tasks based on structured data, called StructGPT.In this framework, we construct the specialized interfaces to collect relevant evidence from structured data (i.e., reading), and let LLMs concentrate on the reasoning task based on the collected information (i.e., reasoning).Specially, we propose an invokinglinearization-generation procedure to support LLMs in reasoning on the structured data with the help of the interfaces.By iterating this procedure with provided interfaces, our approach can gradually approach the target answers to a given query.Experiments conducted on three types of structured data show that StructGPT greatly improves the performance of LLMs, under the few-shot and zero-shot settings.Our codes and data are publicly available at https://github.com/RUCAIBox/StructGPT.