moorcheh/memanto-locomo-results
收藏资源简介:
该数据集包含Memanto在LoCoMo长期对话记忆基准测试中的完整每问题评估结果,涵盖了一个渐进消融研究的所有五个阶段。数据集用于问答任务,涉及长期对话记忆和信息检索增强生成。Memanto是一种用于代理AI的通用记忆层,采用纯向量架构,在LoCoMo基准测试中达到了87.1%的最先进准确率。数据集结构包括五个消融研究阶段的分割,每个分割对应一个评估问题,包含问题ID、类别、问题、预测答案、真实答案、正确性分数和评分理由等字段。
This dataset contains the full per-question evaluation results of Memanto on the LoCoMo long-term conversational memory benchmark, covering all five stages of a progressive ablation study. It is used for question-answering tasks related to long-term conversational memory and retrieval-augmented generation. Memanto is a universal memory layer for agentic AI that achieves state-of-the-art accuracy on LoCoMo (87.1%) using a vector-only architecture. The dataset structure includes five ablation study splits, each corresponding to an evaluated question with fields such as sample_id, category, question, prediction, ground_truth, score, and reasoning.




