Automatic prompt optimization has recently emerged as a strategy for improving the quality of prompts used in Large Language Models (LLMs), with the goal of generating more accurate and useful responses. However, most prior work focuses on direct prompt refinement or model fine-tuning, overlooking the potential of leveraging LLMs' inherent reasoning capability to learn from contrasting examples. In this paper, we present Contrastive Reasoning Prompt Optimization (CRPO), a novel framework that formulates prompt optimization as a retrieval-augmented reasoning process. Our approach retrieves top k reference prompt-response pairs from the HelpSteer2 dataset, an open source collection where each response is annotated for helpfulness, correctness, coherence, complexity, and verbosity, and constructs two complementary optimization paradigms: (1) tiered contrastive reasoning, where the LLM compares high-, medium-, and low-quality exemplars (both prompts and responses) to refine its own generation through reflective reasoning, and (2) multi-metric contrastive reasoning, where the LLM analyzes the best exemplars along each evaluation dimension and integrates their strengths into an optimized prompt. By explicitly contrasting high and low quality exemplars, CRPO enables the model to deduce why certain prompts succeed while others fail, thereby achieving more robust and interpretable optimization. Experimental results on the HelpSteer2 benchmark demonstrate that CRPO significantly outperforms baselines. Our findings highlight the promise of contrastive, retrieval-augmented reasoning for advancing automatic prompt optimization.

Accurate information retrieval (IR) is critical in the financial domain, where investors must identify relevant information from large collections of documents. Traditional IR methods -- whether sparse or dense -- often fall short in retrieval accuracy, as it requires not only capturing semantic similarity but also performing fine-grained reasoning over document structure and domain-specific knowledge. Recent advances in large language models (LLMs) have opened up new opportunities for retrieval with multi-step reasoning, where the model ranks passages through iterative reasoning about which information is most relevant to a given query. However, there exists no benchmark to evaluate such capabilities in the financial domain. To address this gap, we introduce FinAgentBench, the first large-scale benchmark for evaluating retrieval with multi-step reasoning in finance -- a setting we term agentic retrieval. The benchmark consists of 26K expert-annotated examples on S&P-500 listed firms and assesses whether LLM agents can (1) identify the most relevant document type among candidates, and (2) pinpoint the key passage within the selected document. Our evaluation framework explicitly separates these two reasoning steps to address context limitations. This design enables to provide a quantitative basis for understanding retrieval-centric LLM behavior in finance. We evaluate a suite of state-of-the-art models and further demonstrated how targeted fine-tuning can significantly improve agentic retrieval performance. Our benchmark provides a foundation for studying retrieval-centric LLM behavior in complex, domain-specific tasks for finance.

Despite rapid advances in large language models (LLMs), low-resource languages remain excluded from NLP, limiting digital access for millions. We present PunGPT2, the first fully open-source Punjabi generative model suite, trained on a 35GB corpus covering literature, religious texts, news, social discourse, etc. PunGPT2 captures Punjabi's syntactic and morphological richness through a tokenizer optimized for Gurmukhi and Shahmukhi scripts. We introduce Pun-RAG, a retrieval-augmented framework integrating PunGPT2 with a FAISS retriever over a curated Punjabi knowledge base, and Pun-Instruct, an instruction-tuned variant using QLoRA for robust zero-shot summarization, translation, and question answering. Our key innovation, Quantum-RAG, fuses sparse, dense, and quantum kernel embeddings for efficient, context-aware retrieval with low memory overhead, marking the first practical quantum-inspired retrieval in a low-resource LLM. Our models outperform multilingual baselines (mBERT, mT5, MuRIL, BLOOM) on FLORES-200, IndicGenBench, and a new PunjabiEval suite. Quantum-RAG yields +7.4 Recall@10 over FAISS and +3.5 BLEU over mT5 on PunjabiEval. We publicly release all training scripts, hyperparameters, evaluation pipelines, the 35GB Punjabi corpus, the PunjabiEval benchmark, and all model weights, establishing new state-of-the-art results for Punjabi language generation and retrieval.

There is a growing interest in leveraging multiple large language models (LLMs) for automated code optimization. However, industrial platforms deploying multiple LLMs face a critical challenge: prompts optimized for one LLM often fail with others, requiring expensive model-specific prompt engineering. This cross-model prompt engineering bottleneck severely limits the practical deployment of multi-LLM systems in production environments. We introduce Meta-Prompted Code Optimization (MPCO), a framework that automatically generates high-quality, task-specific prompts across diverse LLMs while maintaining industrial efficiency requirements. MPCO leverages metaprompting to dynamically synthesize context-aware optimization prompts by integrating project metadata, task requirements, and LLM-specific contexts. It is an essential part of the ARTEMIS code optimization platform for automated validation and scaling. Our comprehensive evaluation on five real-world codebases with 366 hours of runtime benchmarking demonstrates MPCO's effectiveness: it achieves overall performance improvements up to 19.06% with the best statistical rank across all systems compared to baseline methods. Analysis shows that 96% of the top-performing optimizations stem from meaningful edits. Through systematic ablation studies and meta-prompter sensitivity analysis, we identify that comprehensive context integration is essential for effective meta-prompting and that major LLMs can serve effectively as meta-prompters, providing actionable insights for industrial practitioners.

While reinforcement learning with verifiable rewards (RLVR) is effective to improve the reasoning ability of large language models (LLMs), its reliance on human-annotated labels leads to the scaling up dilemma, especially for complex tasks. Recent self-rewarding methods investigate a label-free alternative to unlock the reasoning capabilities of LLMs, yet they frequently encounter the non-negligible training collapse issue, as the single-view supervision signal easily forms the self-consistent illusion, yielding the reward hacking. Inspired by the success of self-supervised learning, we propose \textit{Co-rewarding}, a novel self-supervised RL framework that improves training stability by seeking complementary supervision from another views. Specifically, we instantiate Co-rewarding in two ways: (1) \textit{Co-rewarding-I} is a data-side instantiation that derives reward signals from contrastive agreement across semantically analogous questions; and (2) \textit{Co-rewarding-II} is a model-side instantiation that maintains a slowly-updated reference teacher with pseudo labels to realize self-distillation. Intuitively, such instantiations introduce different levels of discrepancy to increase the difficulty of training collapse on trivial reasoning solutions. Empirically, Co-rewarding exhibits stable training across various setups, and outperforms other self-rewarding baselines by $+3.31\%$ improvements on average on multiple mathematical reasoning benchmarks, especially by $+7.49\%$ on Llama-3.2-3B-Instruct. Notably, Co-rewarding reaches or even surpasses RLVR with ground-truth (GT) label in several cases, such as a Pass@$1$ of $94.01\%$ on GSM8K with Qwen3-8B-Base remarkably higher than GT. Our code is publicly available at https://github.com/tmlr-group/Co-rewarding.

Pun generation seeks to creatively modify linguistic elements in text to produce humour or evoke double meanings. It also aims to preserve coherence and contextual appropriateness, making it useful in creative writing and entertainment across various media and contexts. Although pun generation has received considerable attention in computational linguistics, there is currently no dedicated survey that systematically reviews this specific area. To bridge this gap, this paper provides a comprehensive review of pun generation datasets and methods across different stages, including conventional approaches, deep learning techniques, and pre-trained language models. Additionally, we summarise both automated and human evaluation metrics used to assess the quality of pun generation. Finally, we discuss the research challenges and propose promising directions for future work.

In a context of malware analysis, numerous approaches rely on Artificial Intelligence to handle a large volume of data. However, these techniques focus on data view (images, sequences) and not on an expert's view. Noticing this issue, we propose a preprocessing that focuses on expert knowledge to improve malware semantic analysis and result interpretability. We propose a new preprocessing method which creates JSON reports for Portable Executable files. These reports gather features from both static and behavioral analysis, and incorporate packer signature detection, MITRE ATT\&CK and Malware Behavior Catalog (MBC) knowledge. The purpose of this preprocessing is to gather a semantic representation of binary files, understandable by malware analysts, and that can enhance AI models' explainability for malicious files analysis. Using this preprocessing to train a Large Language Model for Malware classification, we achieve a weighted-average F1-score of 0.94 on a complex dataset, representative of market reality.

Large language models (LLMs) excel at generating empathic responses in text-based conversations. But, how reliably do they judge the nuances of empathic communication? We investigate this question by comparing how experts, crowdworkers, and LLMs annotate empathic communication across four evaluative frameworks drawn from psychology, natural language processing, and communications applied to 200 real-world conversations where one speaker shares a personal problem and the other offers support. Drawing on 3,150 expert annotations, 2,844 crowd annotations, and 3,150 LLM annotations, we assess inter-rater reliability between these three annotator groups. We find that expert agreement is high but varies across the frameworks' sub-components depending on their clarity, complexity, and subjectivity. We show that expert agreement offers a more informative benchmark for contextualizing LLM performance than standard classification metrics. Across all four frameworks, LLMs consistently approach this expert level benchmark and exceed the reliability of crowdworkers. These results demonstrate how LLMs, when validated on specific tasks with appropriate benchmarks, can support transparency and oversight in emotionally sensitive applications including their use as conversational companions.

Vision-Language models (VLMs) show impressive abilities to answer questions on visual inputs (e.g., counting objects in an image), yet demonstrate higher accuracies when performing an analogous task on text (e.g., counting words in a text). We investigate this accuracy gap by identifying and comparing the \textit{circuits} - the task-specific computational sub-graphs - in different modalities. We show that while circuits are largely disjoint between modalities, they implement relatively similar functionalities: the differences lie primarily in processing modality-specific data positions (an image or a text sequence). Zooming in on the image data representations, we observe they become aligned with the higher-performing analogous textual representations only towards later layers, too late in processing to effectively influence subsequent positions. To overcome this, we patch the representations of visual data tokens from later layers back into earlier layers. In experiments with multiple tasks and models, this simple intervention closes a third of the performance gap between the modalities, on average. Our analysis sheds light on the multi-modal performance gap in VLMs and suggests a training-free approach for reducing it.

Extensive LLM applications demand efficient structured generations, particularly for LR(1) grammars, to produce outputs in specified formats (e.g., JSON). Existing methods primarily parse LR(1) grammars into a pushdown automaton (PDA), leading to runtime execution overhead for context-dependent token processing, especially inefficient under large inference batches. To address these issues, we propose Pre$^3$ that exploits deterministic pushdown automata (DPDA) to optimize the constrained LLM decoding efficiency. First, by precomputing prefix-conditioned edges during the preprocessing, Pre$^3$ enables ahead-of-time edge analysis and thus makes parallel transition processing possible. Second, by leveraging the prefix-conditioned edges, Pre$^3$ introduces a novel approach that transforms LR(1) transition graphs into DPDA, eliminating the need for runtime path exploration and achieving edge transitions with minimal overhead. Pre$^3$ can be seamlessly integrated into standard LLM inference frameworks, reducing time per output token (TPOT) by up to 40% and increasing throughput by up to 36% in our experiments. Our code is available at https://github.com/ModelTC/lightllm.