arXiv:2511.14130v1 Announce Type: cross 
Abstract: With the rapid progress of large language models (LLMs), financial information retrieval has become a critical industrial application. Extracting task-relevant information from lengthy financial filings is essential for both operational and analytical decision-making. The FinAgentBench dataset formalizes this problem through two tasks: document ranking and chunk ranking. We present PRISM, a training-free framework that integrates refined system prompting, in-context learning (ICL), and a lightweight multi-agent system. Each component is examined extensively to reveal their synergies: prompt engineering provides precise task instructions, ICL supplies semantically relevant few-shot examples, and the multi-agent system models coordinated scoring behaviour. Our best configuration achieves an NDCG@5 of 0.71818 on the restricted validation split. We further demonstrate that PRISM is feasible and robust for production-scale financial retrieval. Its modular, inference-only design makes it practical for real-world use cases. The source code is released at https://bit.ly/prism-ailens.

يتناول المقال PRISM، وهو إطار جديد مصمم لاسترجاع المعلومات المالية، مستفيدًا من التقدم في نماذج اللغة الكبيرة (LLMs). يتناول التحديات المتعلقة باستخراج المعلومات ذات الصلة من الوثائق المالية الطويلة، وهو أمر حاسم لاتخاذ القرارات. يتم تقديم مجموعة بيانات FinAgentBench لتشكيل هذه المشكلة من خلال مهام تصنيف الوثائق وتصنيف الأجزاء. يدمج PRISM بين التوجيه المحسن، والتعلم في السياق (ICL)، ونظام متعدد الوكلاء، محققًا درجة NDCG@5 تبلغ 0.71818، مما يدل على فعاليته في تطبيقات استرجاع المعلومات المالية على نطاق واسع.

El artículo presenta PRISM, un nuevo marco diseñado para la recuperación de información financiera, aprovechando los avances en modelos de lenguaje de gran tamaño (LLMs). Aborda los desafíos de extraer información relevante de documentos financieros extensos, lo cual es crucial para la toma de decisiones. Se introduce el conjunto de datos FinAgentBench para formalizar este problema a través de tareas de clasificación de documentos y fragmentos. PRISM integra un prompting refinado, aprendizaje en contexto (ICL) y un sistema multiagente, logrando un NDCG@5 de 0.71818, lo que demuestra su efectiv…

L'article présente PRISM, un cadre novateur conçu pour la récupération d'informations financières, tirant parti des avancées des grands modèles de langage (LLM). Il aborde les défis liés à l'extraction d'informations pertinentes à partir de documents financiers étendus, ce qui est crucial pour la prise de décision. Le jeu de données FinAgentBench est introduit pour formaliser ce problème à travers des tâches de classement de documents et de morceaux. PRISM intègre un prompting affiné, un apprentissage contextuel (ICL) et un système multi-agents, atteignant un score NDCG@5 de 0,71818, démontran…

The article discusses PRISM, a novel framework designed for financial information retrieval, leveraging advancements in large language models (LLMs). It addresses the challenges of extracting relevant information from extensive financial documents, which is crucial for decision-making. The FinAgentBench dataset is introduced to formalize this issue through document and chunk ranking tasks. PRISM integrates refined prompting, in-context learning (ICL), and a multi-agent system, achieving an NDCG@5 score of 0.71818, demonstrating its effectiveness for large-scale financial retrieval applications…

PRISM: Prompt-Refined In-Context System Modelling for Financial Retrieval

arXiv:2502.12119v3 Announce Type: replace 
Abstract: Visual instruction tuning adapts pre-trained Multimodal Large Language Models (MLLMs) to follow human instructions for real-world applications. However, the rapid growth of these datasets introduces significant redundancy, leading to increased computational costs. Existing methods for selecting instruction data aim to prune this redundancy, but predominantly rely on computationally demanding techniques such as proxy-based inference or training-based metrics. Consequently, the substantial computational costs incurred by these selection processes often exacerbate the very efficiency bottlenecks they are intended to resolve, posing a significant challenge to the scalable and effective tuning of MLLMs. To address this challenge, we first identify a critical, yet previously overlooked, factor: the anisotropy inherent in visual feature distributions. We find that this anisotropy induces a \textit{Global Semantic Drift}, and overlooking this phenomenon is a key factor limiting the efficiency of current data selection methods. Motivated by this insight, we devise \textbf{PRISM}, the first training-free framework for efficient visual instruction selection. PRISM surgically removes the corrupting influence of global background features by modeling the intrinsic visual semantics via implicit re-centering. Empirically, PRISM reduces the end-to-end time for data selection and model tuning to just 30\% of conventional pipelines. More remarkably, it achieves this efficiency while simultaneously enhancing performance, surpassing models fine-tuned on the full dataset across eight multimodal and three language understanding benchmarks, culminating in a 101.7\% relative improvement over the baseline. The code is available for access via \href{https://github.com/bibisbar/PRISM}{this repository}.

تم تقديم طريقة جديدة تُدعى PRISM لتحسين اختيار بيانات التدريب لنماذج اللغة متعددة الوسائط (MLLMs)، حيث تتناول الازدواجية في مجموعات البيانات التي تنمو بسرعة مما يزيد من التكاليف الحاسوبية. تهدف هذه الطريقة للاختيار الذاتي إلى تحسين الكفاءة دون الحاجة إلى تدريب مكثف أو تقنيات استدلال تعتمد على الوكلاء.

Se ha introducido un nuevo método llamado PRISM para optimizar la selección de datos de entrenamiento para Modelos de Lenguaje Multimodal (MLLMs), abordando la redundancia en conjuntos de datos en rápido crecimiento que aumentan los costos computacionales. Este método de selección intrínseca auto-poda tiene como objetivo mejorar la eficiencia sin la necesidad de entrenamiento extenso o técnicas de inferencia basadas en proxies.

Une nouvelle méthode appelée PRISM a été introduite pour optimiser la sélection des données d'entraînement pour les modèles de langage multimodaux (MLLMs), en s'attaquant à la redondance dans des ensembles de données en forte croissance qui augmentent les coûts computationnels. Cette méthode de sélection intrinsèque auto-élagante vise à améliorer l'efficacité sans nécessiter d'entraînement extensif ou de techniques d'inférence basées sur des proxies.

A new method called PRISM has been introduced to optimize the selection of training data for Multimodal Large Language Models (MLLMs), addressing the redundancy in rapidly growing datasets that increases computational costs. This self-pruning intrinsic selection method aims to enhance efficiency without the need for extensive training or proxy-based inference techniques.

PRISM: Self-Pruning Intrinsic Selection Method for Training-Free Multimodal Data Selection

One More Thing in AI – Your Shortcut to AI Mastery

PRISM: Prompt-Refined In-Context System Modelling for Financial Retrieval

Was this article worth reading? Share it

One More Thing in AI

finlight.me

LucidQuery AI

Airparser

MyFramework

Supametas.AI

Ready to build your own newsroom?