{"@context":"https://schema.org","@type":"NewsArticle","generatedAt":"2026-10-06T08:00:48.640Z","headline":"NVIDIA 发布开源表格基础模型 Kumo Tabular，在 TabArena 等四项基准排名第一","description":"NVIDIA 发布开源表格基础模型 Kumo Tabular，对带标签表格做单次前向推理即可完成分类和回归，无需训练、调参或特征工程。","url":"https://www.aioga.com/news/tcf79upi169og7ftecq28pps2/","mainEntityOfPage":"https://www.aioga.com/news/tcf79upi169og7ftecq28pps2/","datePublished":"2026-09-29T15:30:38.000Z","dateModified":"2026-09-29T15:30:38.000Z","inLanguage":"zh-CN","publisher":{"@type":"NewsMediaOrganization","name":"Aioga","url":"https://www.aioga.com"},"citation":["https://huggingface.co/blog/nvidia/kumo-tabular","https://aihot.news/items/tcf79upi169og7ftecq28pps2"],"canonicalUrl":"https://www.aioga.com/news/tcf79upi169og7ftecq28pps2/","directAnswer":{"@type":"Answer","text":"NVIDIA 发布开源表格基础模型 Kumo Tabular，面向表格分类与回归。来源称，模型可依据带标签的表格，在单次前向推理中预测新行标签，无需训练、调参或特征工程；并在 TabArena、BeyondArena、TALENT 和 ScoringBench 四项基准排名第一。","url":"https://www.aioga.com/news/tcf79upi169og7ftecq28pps2/","dateCreated":"2026-09-29T15:30:38.000Z","author":{"@type":"Organization","@id":"https://www.aioga.com/authors/aioga-editorial/#editorial-team","name":"Aioga Editorial Team","url":"https://www.aioga.com/authors/aioga-editorial/"}},"evidence":[{"@type":"CreativeWork","name":"huggingface.co source article","url":"https://huggingface.co/blog/nvidia/kumo-tabular","datePublished":"2026-09-29T15:30:38.000Z","provider":{"@type":"Organization","name":"huggingface.co","url":"https://huggingface.co/blog/nvidia/kumo-tabular"}},{"@type":"CreativeWork","name":"AIHot archive record","url":"https://aihot.news/items/tcf79upi169og7ftecq28pps2","datePublished":"2026-09-29T15:30:38.000Z","provider":{"@type":"Organization","name":"AIHot","url":"https://aihot.news/items/tcf79upi169og7ftecq28pps2"}}],"aggregationSource":"Hugging Face：Blog（RSS）","originalPublisher":{"name":"huggingface.co","url":"https://huggingface.co/blog/nvidia/kumo-tabular"},"geoDeepAnswer":null,"article":{"id":"tcf79upi169og7ftecq28pps2","slug":"tcf79upi169og7ftecq28pps2","url":"https://www.aioga.com/news/tcf79upi169og7ftecq28pps2/","title":"NVIDIA 发布开源表格基础模型 Kumo Tabular，在 TabArena 等四项基准排名第一","title_en":"","summary":"NVIDIA 发布开源表格基础模型 Kumo Tabular，对带标签表格做单次前向推理即可完成分类和回归，无需训练、调参或特征工程。","source":"Hugging Face：Blog（RSS）","sourceUrl":"https://huggingface.co/blog/nvidia/kumo-tabular","aiHotUrl":"https://aihot.news/items/tcf79upi169og7ftecq28pps2","publishedAt":"2026-09-29T15:30:38.000Z","category":"行业动态","score":72,"selected":true,"articleBody":["NVIDIA Kumo Tabular, part of the NVIDIA Kumo Structured model collection, is an open foundation model for tabular data now available on Hugging Face：https://huggingface.co/nvidia/Kumo-Tabular. Given a table of labeled rows, it predicts the labels of new rows in a single forward pass, with no training, no tuning, and no feature engineering, for both classification and regression. It was pretrained only on artificial data, comes in three sizes (28M to 215M parameters), runs through our open-source library：https://github.com/NVIDIA/structured-data-models, and is released under the OpenMDW-1.1 license：https://openmdw.ai/license/1-1/ for commercial use. It ranks first on the four benchmarks TabArena：https://github.com/autogluon/tabarena, BeyondArena：https://github.com/autogluon/tabarena, TALENT：https://github.com/LAMDA-Tabular/TALENT and ScoringBench：https://github.com/jonaslandsgesell/ScoringBench.","Tabular data is the backbone of enterprise machine learning. Customer records, transactions, sensor logs, claims, and orders all live in tables, and predicting churn, default, demand, or price from them is among the most common machine learning tasks in industry. For two decades, this work has been done with gradient-boosted trees, and it has worked well. But the lifecycle around those models has barely changed. Every new question means collecting labels, engineering features, searching hyperparameters, validating, and deploying a model that knows nothing about tables in general and learns each task from scratch.","Large Language Models showed a different way of working with new tasks. Given a few examples in the prompt, a pretrained model solves the task without updating a single weight. This is in-context learning , and it applies to tables just as well as to text: a model pretrained on millions of tables can read a labeled table as its context and predict the labels of new rows directly.","Today, we are releasing NVIDIA Kumo Tabular (GitHub：https://github.com/NVIDIA/structured-data-models, HuggingFace：https://huggingface.co/nvidia/Kumo-Tabular), an open foundation model for tabular classification and regression. Given a table with labeled rows and the rows you want predictions for, Kumo Tabular returns class probabilities or numeric predictions in a single forward pass.","Kumo Tabular is a Transformer built around the structure of a table, utilizing column, row and in-context attention as introduced in TabICL：https://github.com/soda-inria/tabicl and TabPFN：https://github.com/PriorLabs/tabpfn. To predict a label it has to do three things: (1) understand what each value means within its column, (2) understand how the columns of a row interact, and (3) relate the context rows with existing labels to the query rows with unknown labels. Kumo Tabular achieves this as follows:","：https://cdn-uploads.huggingface.co/production/uploads/684040a5de1ad7f4fcec9508/TkT695UAWJnUIyEydIYRv.png","Cell Embedding: A group of cells becomes a token. Numerical and categorical values pass through Fourier features, sines and cosines of learned frequencies, with separate weights for each type. Missing values need no imputation and are treated specially. Finally, every token in the context receives a label embedding.","Row Embedding: We then turn each row into an embedding by alternating two kinds of attention multiple times. Column attention looks down a single column and learns what a value means in the distribution of its column, e.g. , whether a 42 is typical or extreme, via induced self-attention. Its cost therefore grows linearly with the number of rows. Row attention looks across the tokens of a single row and learns how features interact, with rotary positions to tell columns apart. Four learnable [CLS] tokens join each row and act as the final readout of a row. After this row compression, the cost of the final stage no longer depends on the number of columns.","In-context Learning: A final Transformer operates on the row embeddings. Context rows attend to each other, while query rows attend to context rows only. Each prediction therefore depends only on the context and on the row itself, not on which other rows are scored alongside it. Because the context never looks at the queries, its keys and values are computed once and can be reused for follow-up predictions. Query rows utilize Test-GQA, which shrinks the cache that every prediction reads. A head turns each query row into class probabilities for classification and 999 quantiles for regression, from which a point prediction and an uncertainty estimate follow.","Length-aware Attention Temperature: Softmax attention spreads out as the number of keys grows. Attention that is sharp over a few hundred rows can dissolve over tens of thousands, which is exactly the situation when a table at inference is much larger than a typical training table. Kumo Tabular therefore scales every query by a temperature that grows with the logarithm of the number of keys, with a coefficient learned separately for each attention head. The result is attention that stays sharp as tables grow longer or wider.","Kumo Tabular is pretrained entirely on artificial tables. Each training table is sampled from a Structural Causal Model (SCM) in the six steps shown below:","：https://cdn-uploads.huggingface.co/production/uploads/684040a5de1ad7f4fcec9508/ipJNJTNO5ZH2zviIdR8eo.png","We first draw a configuration for the whole table, from its size and task to its mechanisms and missingness. A random causal graph then links hidden variables, evaluated from root to leaf via randomly drawn functions at every node ( e.g. , linear maps, small neural networks, trees or Gaussian processes). Some nodes become numerical or categorical columns, one becomes the target, and the rest stay hidden, like the unmeasured causes behind real data. Post-processing correlates groups of columns, clips outliers, and injects missing values, and a quick tree-ensemble check discards any table without a learnable signal. Because the generator is a procedural sampler rather than a trained model, it produces an endless supply of tables, each with a new graph and new mechanisms.","Real-world tables are messy, so we built more of their imperfections into the generator. Values go missing in several patterns, some features are coarsened so that duplicate rows may disagree on their label, some categorical columns carry many levels, and regression targets can be heavy-tailed. A model that has seen millions of such tables learns to handle these imperfections without any cleanup.","On every artificial table, the model sees most of the rows with their labels as context and learns to predict the labels of the remaining rows, with a cross-entropy loss for classification and a quantile loss for regression. Classification and regression are trained as separate models. Similarly to TabICLv2, training runs in three stages. The first and longest stage uses tables of 1,024 rows and up to 100 columns and teaches the model what tables look like. The second stage varies the context from 400 to 10,240 rows, and the third extends it to 60,000 rows, still with up to 100 columns. In total, Kumo Tabular-Small/Medium/Large saw about 35/71/137 million artificial tables.","Our training recipe and artificial data generators will be released soon.","We ran all three Kumo Tabular sizes with default settings against the full TabArena：https://github.com/autogluon/tabarena leaderboard, spanning tuned gradient-boosted trees, AutoGluon, and the latest tabular foundation models. Kumo Tabular ranks first overall with an ELO of 1950 while running 17 faster than LimiX-2：https://github.com/limix-ldm-ai/LimiX under a uniform single RTX 6000 Pro evaluation setup. Across all three three model sizes, Kumo Tabular establishes a new state-of-the-art on the accuracy-efficiency Pareto front:","：https://cdn-uploads.huggingface.co/production/uploads/684040a5de1ad7f4fcec9508/CpOXm3g9du6Uw3zQDh2QC.png","We also evaluated Kumo Tabular on BeyondArena：https://github.com/autogluon/tabarena, TALENT：https://github.com/LAMDA-Tabular/TALENT and ScoringBench：https://github.com/jonaslandsgesell/ScoringBench. On BeyondArena, Kumo Tabular reaches an ELO of 1418 with an Improvability score of 7.78%, placing first on the leaderboard. On TALENT, it achieves the top overall ranking across classification accuracy, classification log-loss, and regression RMSE, with average ranks of 6.67, 3.98, and 4.22. On ScoringBench, a benchmark for predictive distributions, Kumo Tabular-Large and Medium rank first and second on average rank.","Kumo Tabular works on numerical and categorical columns only, while text, images, or timestamps can be turned into features via built-in pre-processing recipes. A single forward pass covers up to 10 classes, which the library extends to any number of classes with error-correcting output codes. Accuracy may degrade on tables far beyond the training ranges or when the query rows come from a different distribution than the context rows, so, as with any predictive model, validate accuracy and calibration on your own held-out data before deployment.","Kumo Tabular runs via NVIDIA's newly released GPU-native library for structured-data-models ：https://github.com/NVIDIA/structured-data-models. The library downloads the weights from the Hub on first use and provides the preprocessing, ensembling, and many-class handling used in our evaluations. The code below is all it takes to go from a pandas.DataFrame to a prediction:","Kumo Tabular is released under the OpenMDW License Agreement, version 1.1：/blog/nvidia/(https://openmdw.ai/license/1-1/. NVIDIA believes Trustworthy AI is a shared responsibility, and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their supporting model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse. Please report model quality, risk, security vulnerabilities, or NVIDIA AI concerns here：https://github.com/NVIDIA/structured-data-models/issues.","We thank David Holzmüller：https://dholzmueller.github.io/ for contributing significant ideas and ablations to Kumo Tabular. We thank Vignesh Kothapalli：https://kvignesh1420.github.io/ for his help on Kumo Tabular during his internship."],"articleImages":[{"sourceUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/640ef81aa92fedb0e84ec097/LHbb4-MfvQAor4La_9cqO.jpeg","alt":"","afterParagraph":0,"url":"/media/articles/tcf79upi169og7ftecq28pps2/37a96aab25891725.webp"},{"sourceUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6ab4e1d4dc335892cb51ddc0/oQryM5i2Oo_phA8rqgV4b.jpeg","alt":"","afterParagraph":0,"url":"/media/articles/tcf79upi169og7ftecq28pps2/964b3bf9ba6c1bea.webp"},{"sourceUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6a96adc7d1fd32fbb3bac0ce/TLWZJYbn9-ljfqYNL8LPH.png","alt":"","afterParagraph":0,"url":"/media/articles/tcf79upi169og7ftecq28pps2/ad35914d7b869573.webp"},{"sourceUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1669189789447-629f3b18ee05727ce328ccbe.jpeg","alt":"","afterParagraph":0,"url":"/media/articles/tcf79upi169og7ftecq28pps2/0a69297c7952744b.webp"}],"mediaStatus":"ok","articleBodyZh":["NVIDIA Kumo Tabular，是 NVIDIA Kumo Structured 模型集合的一部分，是一个用于表格数据的开放基础模型，目前可在 Hugging Face 上获取：https://huggingface.co/nvidia/Kumo-Tabular。给定包含标签的表格行，它可以在一次前向传播中预测新行的标签，无需训练、无需调优，也无需特征工程，可用于分类和回归。它仅在人工数据上进行了预训练，提供三种规模（从 28M 到 215M 参数），可通过我们的开源库运行：https://github.com/NVIDIA/structured-data-models，并在 OpenMDW-1.1 许可证下发布：https://openmdw.ai/license/1-1/，可用于商业用途。在四个基准测试中排名第一：TabArena：https://github.com/autogluon/tabarena、BeyondArena：https://github.com/autogluon/tabarena、TALENT：https://github.com/LAMDA-Tabular/TALENT 和 ScoringBench：https://github.com/jonaslandsgesell/ScoringBench。","表格数据是企业机器学习的核心。客户记录、交易、传感器日志、理赔和订单都存储在表格中，从中预测客户流失、违约、需求或价格是行业内最常见的机器学习任务之一。二十年来，这项工作一直依赖梯度提升树，并取得了良好效果。但这些模型的生命周期几乎没有变化。每遇到一个新问题，都意味着需要收集标签、工程特征、搜索超参数、验证并部署一个对表格一无所知、从零学习每项任务的模型。","大型语言模型展示了处理新任务的不同方式。给出提示中的几个示例，预训练模型无需更新任何权重即可解决任务。这就是上下文学习，它同样适用于表格和文本：一个在数百万表格上预训练的模型可以将带标签的表格作为上下文，直接预测新行的标签。","今天，我们发布了 NVIDIA Kumo Tabular（GitHub：https://github.com/NVIDIA/structured-data-models，HuggingFace：https://huggingface.co/nvidia/Kumo-Tabular），这是一个用于表格分类和回归的开放基础模型。给定带标签的表格行及需预测的行，Kumo Tabular 可以在一次前向传播中返回类别概率或数值预测。","Kumo Tabular 是一个基于表格结构构建的变换器，利用列、行和上下文关注，这一功能在 TabICL：https：//github.com/soda-inria/tabicl 和 TabPFN：https：//github.com/PriorLabs/tabpfn 中引入。要预测标签，它需要做三件事：（1） 理解每个值在其列中的含义，（2） 理解行列之间的相互作用，（3） 将上下文行与已有标签的上下文行与未知标签的查询行关联起来。Kumo Tabular 的实现方式如下：","：https://cdn-uploads.huggingface.co/production/uploads/684040a5de1ad7f4fcec9508/TkT695UAWJnUIyEydIYRv.png","单元嵌入：一组单元格成为一个符号。数值和类别值通过傅里叶特征、学习频率的正弦和余弦，每种类型有不同的权重。缺失值无需补补，且会被特别处理。最后，上下文中的每个标记都获得标签嵌入。","行嵌入：然后我们通过交替使用两种注意力多次，将每行转化为嵌入。列注意力通过引导自注意，向下观察单一列，了解该值在列分布中的含义，例如42是典型还是极端。因此，其成本随行数线性增长。行注意力会跨越单行的标记，学习特征之间的相互作用，并通过旋转位置区分列。每行加入四个可学习的[CLS]标记，作为行的最终读出。经过行压缩后，最终阶段的成本不再依赖于列数。","上下文学习：最后一个变换器对行嵌入进行操作。上下文行相互关注，而查询行只关注上下文行。因此，每个预测仅依赖上下文和行本身，而非与其他行并列评分的行。由于上下文从不查看查询，其键和值只计算一次，可重复用于后续预测。查询行利用Test-GQA，缩小每个预测读取的缓存。一个头将每个查询行转换为分类类概率和999分位数进行回归，基于此得出点预测和不确定性估计。","长度感知注意力温度：随着键的数量增加，Softmax 注意力会扩散开来。在几百行上非常集中的注意力，在几万行上可能会消散，这正是推理时表格比典型训练表格大得多时的情况。因此，Kumo Tabular 会根据键的数量的对数增加温度来缩放每个查询，并且每个注意力头都有单独学习的系数。其结果是，随着表格长度或宽度增加，注意力仍然保持集中。","Kumo Tabular 完全在人工表格上进行预训练。每个训练表格都是从结构因果模型（SCM）中按以下六个步骤采样的：","：https://cdn-uploads.huggingface.co/production/uploads/684040a5de1ad7f4fcec9508/ipJNJTNO5ZH2zviIdR8eo.png","我们首先为整个表格绘制一个配置，包括其大小、任务、机制和缺失情况。然后，随机因果图将隐藏变量连接起来，通过在每个节点随机抽取的函数（例如线性映射、小型神经网络、树或高斯过程）从根到叶进行评估。一些节点成为数值或类别列，其中一列成为目标，其余的保持隐藏，就像真实数据背后的未测量因果。后处理步骤会关联列组、剪裁异常值并注入缺失值，同时通过快速的树集合检查丢弃任何没有可学习信号的表格。因为生成器是程序化采样器而非训练模型，它可以生成无穷尽的表格，每个表格都有新的图和新的机制。","真实世界的表格很混乱，因此我们在生成器中增加了更多此类不完善之处。值缺失会呈现多种模式，一些特征被粗化，使得重复行可能在标签上不一致，一些类别列包含许多级别，回归目标可能呈现重尾分布。一个见过数百万个此类表格的模型会学会在无需任何清理的情况下处理这些不完善之处。","在每个人工表格上，模型将大部分带有标签的行作为上下文，并学习预测剩余行的标签，对于分类任务使用交叉熵损失，对于回归任务使用分位数损失。分类和回归作为独立模型进行训练。与 TabICLv2 类似，训练分为三个阶段进行。第一阶段最长，使用 1,024 行、最多 100 列的表格，教模型理解表格的样子。第二阶段将上下文行数从 400 变动至 10,240 行，第三阶段扩展至 60,000 行，仍然最多 100 列。总的来说，Kumo Tabular-Small/Medium/Large 分别看到约 3,500 万 / 7,100 万 / 1.37 亿个人工表格。","我们的训练方法和人工数据生成器将很快发布。","我们使用默认设置在完整的 TabArena：https://github.com/autogluon/tabarena 排行榜上运行了三种 Kumo Tabular 尺寸，涵盖经过调优的梯度提升树、AutoGluon 以及最新的表格基础模型。Kumo Tabular 总体排名第一，ELO 为 1950，在统一的单 RTX 6000 Pro 测评环境下，比 LimiX-2：https://github.com/limix-ldm-ai/LimiX 快 17 倍。在三种模型尺寸中，Kumo Tabular 在准确性-效率帕累托前沿上建立了新的最先进水平：","：https://cdn-uploads.huggingface.co/production/uploads/684040a5de1ad7f4fcec9508/CpOXm3g9du6Uw3zQDh2QC.png","我们还在 BeyondArena：https://github.com/autogluon/tabarena、TALENT：https://github.com/LAMDA-Tabular/TALENT 和 ScoringBench：https://github.com/jonaslandsgesell/ScoringBench 上评估了 Kumo Tabular。在 BeyondArena 上，Kumo Tabular 的 ELO 为 1418，改进分数为 7.78%，位列排行榜第一。在 TALENT 上，它在分类准确率、分类对数损失和回归均方根误差上获得整体最高排名，平均排名分别为 6.67、3.98 和 4.22。在用于预测分布的 ScoringBench 中，Kumo Tabular-Large 和 Medium 在平均排名上分别位列第一和第二。","Kumo Tabular 仅适用于数值列和类别列，而文本、图像或时间戳可以通过内置预处理方案转换为特征。单次前向传播最多涵盖 10 个类别，该库通过纠错输出编码扩展到任意数量的类别。当表格超出训练范围很远或查询行与上下文行来自不同分布时，准确率可能下降，因此，与任何预测模型一样，在部署前应在自己的保留数据上验证准确性和校准性。","Kumo Tabular 运行于 NVIDIA 新发布的适用于结构化数据模型的 GPU 原生库上：https://github.com/NVIDIA/structured-data-models。该库在首次使用时会从 Hub 下载权重，并提供我们在评估中使用的预处理、集成和多类别处理功能。以下代码即可将 pandas.DataFrame 转换为预测结果：","Kumo Tabular 依据 OpenMDW 许可协议 1.1 版本发布：/blog/nvidia/(https://openmdw.ai/license/1-1/)。NVIDIA 认为值得信赖的 AI 是共同责任，我们已建立政策和实践以支持广泛的 AI 应用开发。遵循服务条款下载或使用时，开发者应与其所支持的模型团队协作，以确保该模型满足相关行业和使用案例的要求，并防范不可预见的产品滥用。请在此报告模型质量、风险、安全漏洞或 NVIDIA AI 相关问题：https://github.com/NVIDIA/structured-data-models/issues。","我们感谢 David Holzmüller：https://dholzmueller.github.io/ 对 Kumo Tabular 提供的重要想法和消融实验。我们感谢 Vignesh Kothapalli：https://kvignesh1420.github.io/ 在实习期间对 Kumo Tabular 的帮助。"],"translationStatus":"translated","bodyOrigin":"source-page","editorial":{"summary":"NVIDIA 发布开源表格基础模型 Kumo Tabular，面向表格分类与回归。来源称，模型可依据带标签的表格，在单次前向推理中预测新行标签，无需训练、调参或特征工程；并在 TabArena、BeyondArena、TALENT 和 ScoringBench 四项基准排名第一。","background":"Kumo Tabular 属于 NVIDIA Kumo Structured 模型系列，采用围绕表格结构设计的 Transformer。来源称其仅使用人工数据预训练，提供 28M 至 215M 参数的三个版本，通过开源库运行，并以 OpenMDW-1.1 许可发布，可用于商业用途。","viewpoint":"Aioga 判断：这项发布将表格任务的重点放在预训练模型直接处理带标签上下文的方式上。基准排名是来源报告的结果，但现有材料没有提供各项测试的具体条件，不能据此推断模型在所有实际数据集上都占优。","implications":"可能影响：免训练推理的设计或值得表格机器学习团队评估，但来源未说明其在特定业务数据上的效果、成本或部署表现。采用前需要结合自身数据验证，并核对许可条款；基准排名不代表对所有任务的效果保证。","nextStep":"后续观察：可关注模型仓库、开源库及四项基准的公开材料，核实评测设置、复现结果与适用限制；若开展试用，建议记录本地数据上的分类和回归表现，并确认其与现有流程的兼容性。","evidenceRefs":["title","summary","articleBody","source"],"status":"published","aiGenerated":true,"autoApproved":true,"generatedBy":"aioga-editorial:gpt-5.6-sol","reviewedBy":"aioga-editorial-review:gpt-5.6-sol","generatedAt":"2026-09-30T06:15:21.724Z","sourceHash":"81c2d002bdd514fc","review":{"approved":true,"groundedness":97,"clarity":94,"duplicationRisk":18,"blockingIssues":[],"notes":["“可用于商业用途”应以 OpenMDW-1.1 许可的完整条款为准，候选内容已保留核对许可条款的提示。","“将表格任务的重点放在……”属于明确标注的编辑判断，不构成事实冒充。","关于基准测试条件、实际数据集表现、成本和部署表现的表述与来源范围一致，未作过度推断。"]},"validation":{"passed":true,"mode":"ai-auto","revisions":0,"checks":["schema","length","source-attribution","editorial-labels","inference-boundary","low-source-overlap","no-html","independent-ai-review"]}},"tags":["行业动态","Hugging Face：Blog（RSS）"],"translations":{"zh-CN":{"title":"NVIDIA 发布开源表格基础模型 Kumo Tabular，在 TabArena 等四项基准排名第一","summary":"NVIDIA 发布开源表格基础模型 Kumo Tabular，对带标签表格做单次前向推理即可完成分类和回归，无需训练、调参或特征工程。","category":"行业动态","source":"huggingface.co","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"NVIDIA 发布开源表格基础模型 Kumo Tabular，在 TabArena 等四项基准排名第一 - Aioga AI资讯","description":"NVIDIA 发布开源表格基础模型 Kumo Tabular，对带标签表格做单次前向推理即可完成分类和回归，无需训练、调参或特征工程。","url":"https://www.aioga.com/news/tcf79upi169og7ftecq28pps2/","articleBody":["NVIDIA Kumo Tabular，是 NVIDIA Kumo Structured 模型集合的一部分，是一个用于表格数据的开放基础模型，目前可在 Hugging Face 上获取：https://huggingface.co/nvidia/Kumo-Tabular。给定包含标签的表格行，它可以在一次前向传播中预测新行的标签，无需训练、无需调优，也无需特征工程，可用于分类和回归。它仅在人工数据上进行了预训练，提供三种规模（从 28M 到 215M 参数），可通过我们的开源库运行：https://github.com/NVIDIA/structured-data-models，并在 OpenMDW-1.1 许可证下发布：https://openmdw.ai/license/1-1/，可用于商业用途。在四个基准测试中排名第一：TabArena：https://github.com/autogluon/tabarena、BeyondArena：https://github.com/autogluon/tabarena、TALENT：https://github.com/LAMDA-Tabular/TALENT 和 ScoringBench：https://github.com/jonaslandsgesell/ScoringBench。","表格数据是企业机器学习的核心。客户记录、交易、传感器日志、理赔和订单都存储在表格中，从中预测客户流失、违约、需求或价格是行业内最常见的机器学习任务之一。二十年来，这项工作一直依赖梯度提升树，并取得了良好效果。但这些模型的生命周期几乎没有变化。每遇到一个新问题，都意味着需要收集标签、工程特征、搜索超参数、验证并部署一个对表格一无所知、从零学习每项任务的模型。","大型语言模型展示了处理新任务的不同方式。给出提示中的几个示例，预训练模型无需更新任何权重即可解决任务。这就是上下文学习，它同样适用于表格和文本：一个在数百万表格上预训练的模型可以将带标签的表格作为上下文，直接预测新行的标签。","今天，我们发布了 NVIDIA Kumo Tabular（GitHub：https://github.com/NVIDIA/structured-data-models，HuggingFace：https://huggingface.co/nvidia/Kumo-Tabular），这是一个用于表格分类和回归的开放基础模型。给定带标签的表格行及需预测的行，Kumo Tabular 可以在一次前向传播中返回类别概率或数值预测。","Kumo Tabular 是一个基于表格结构构建的变换器，利用列、行和上下文关注，这一功能在 TabICL：https：//github.com/soda-inria/tabicl 和 TabPFN：https：//github.com/PriorLabs/tabpfn 中引入。要预测标签，它需要做三件事：（1） 理解每个值在其列中的含义，（2） 理解行列之间的相互作用，（3） 将上下文行与已有标签的上下文行与未知标签的查询行关联起来。Kumo Tabular 的实现方式如下：","：https://cdn-uploads.huggingface.co/production/uploads/684040a5de1ad7f4fcec9508/TkT695UAWJnUIyEydIYRv.png","单元嵌入：一组单元格成为一个符号。数值和类别值通过傅里叶特征、学习频率的正弦和余弦，每种类型有不同的权重。缺失值无需补补，且会被特别处理。最后，上下文中的每个标记都获得标签嵌入。","行嵌入：然后我们通过交替使用两种注意力多次，将每行转化为嵌入。列注意力通过引导自注意，向下观察单一列，了解该值在列分布中的含义，例如42是典型还是极端。因此，其成本随行数线性增长。行注意力会跨越单行的标记，学习特征之间的相互作用，并通过旋转位置区分列。每行加入四个可学习的[CLS]标记，作为行的最终读出。经过行压缩后，最终阶段的成本不再依赖于列数。","上下文学习：最后一个变换器对行嵌入进行操作。上下文行相互关注，而查询行只关注上下文行。因此，每个预测仅依赖上下文和行本身，而非与其他行并列评分的行。由于上下文从不查看查询，其键和值只计算一次，可重复用于后续预测。查询行利用Test-GQA，缩小每个预测读取的缓存。一个头将每个查询行转换为分类类概率和999分位数进行回归，基于此得出点预测和不确定性估计。","长度感知注意力温度：随着键的数量增加，Softmax 注意力会扩散开来。在几百行上非常集中的注意力，在几万行上可能会消散，这正是推理时表格比典型训练表格大得多时的情况。因此，Kumo Tabular 会根据键的数量的对数增加温度来缩放每个查询，并且每个注意力头都有单独学习的系数。其结果是，随着表格长度或宽度增加，注意力仍然保持集中。","Kumo Tabular 完全在人工表格上进行预训练。每个训练表格都是从结构因果模型（SCM）中按以下六个步骤采样的：","：https://cdn-uploads.huggingface.co/production/uploads/684040a5de1ad7f4fcec9508/ipJNJTNO5ZH2zviIdR8eo.png","我们首先为整个表格绘制一个配置，包括其大小、任务、机制和缺失情况。然后，随机因果图将隐藏变量连接起来，通过在每个节点随机抽取的函数（例如线性映射、小型神经网络、树或高斯过程）从根到叶进行评估。一些节点成为数值或类别列，其中一列成为目标，其余的保持隐藏，就像真实数据背后的未测量因果。后处理步骤会关联列组、剪裁异常值并注入缺失值，同时通过快速的树集合检查丢弃任何没有可学习信号的表格。因为生成器是程序化采样器而非训练模型，它可以生成无穷尽的表格，每个表格都有新的图和新的机制。","真实世界的表格很混乱，因此我们在生成器中增加了更多此类不完善之处。值缺失会呈现多种模式，一些特征被粗化，使得重复行可能在标签上不一致，一些类别列包含许多级别，回归目标可能呈现重尾分布。一个见过数百万个此类表格的模型会学会在无需任何清理的情况下处理这些不完善之处。","在每个人工表格上，模型将大部分带有标签的行作为上下文，并学习预测剩余行的标签，对于分类任务使用交叉熵损失，对于回归任务使用分位数损失。分类和回归作为独立模型进行训练。与 TabICLv2 类似，训练分为三个阶段进行。第一阶段最长，使用 1,024 行、最多 100 列的表格，教模型理解表格的样子。第二阶段将上下文行数从 400 变动至 10,240 行，第三阶段扩展至 60,000 行，仍然最多 100 列。总的来说，Kumo Tabular-Small/Medium/Large 分别看到约 3,500 万 / 7,100 万 / 1.37 亿个人工表格。","我们的训练方法和人工数据生成器将很快发布。","我们使用默认设置在完整的 TabArena：https://github.com/autogluon/tabarena 排行榜上运行了三种 Kumo Tabular 尺寸，涵盖经过调优的梯度提升树、AutoGluon 以及最新的表格基础模型。Kumo Tabular 总体排名第一，ELO 为 1950，在统一的单 RTX 6000 Pro 测评环境下，比 LimiX-2：https://github.com/limix-ldm-ai/LimiX 快 17 倍。在三种模型尺寸中，Kumo Tabular 在准确性-效率帕累托前沿上建立了新的最先进水平：","：https://cdn-uploads.huggingface.co/production/uploads/684040a5de1ad7f4fcec9508/CpOXm3g9du6Uw3zQDh2QC.png","我们还在 BeyondArena：https://github.com/autogluon/tabarena、TALENT：https://github.com/LAMDA-Tabular/TALENT 和 ScoringBench：https://github.com/jonaslandsgesell/ScoringBench 上评估了 Kumo Tabular。在 BeyondArena 上，Kumo Tabular 的 ELO 为 1418，改进分数为 7.78%，位列排行榜第一。在 TALENT 上，它在分类准确率、分类对数损失和回归均方根误差上获得整体最高排名，平均排名分别为 6.67、3.98 和 4.22。在用于预测分布的 ScoringBench 中，Kumo Tabular-Large 和 Medium 在平均排名上分别位列第一和第二。","Kumo Tabular 仅适用于数值列和类别列，而文本、图像或时间戳可以通过内置预处理方案转换为特征。单次前向传播最多涵盖 10 个类别，该库通过纠错输出编码扩展到任意数量的类别。当表格超出训练范围很远或查询行与上下文行来自不同分布时，准确率可能下降，因此，与任何预测模型一样，在部署前应在自己的保留数据上验证准确性和校准性。","Kumo Tabular 运行于 NVIDIA 新发布的适用于结构化数据模型的 GPU 原生库上：https://github.com/NVIDIA/structured-data-models。该库在首次使用时会从 Hub 下载权重，并提供我们在评估中使用的预处理、集成和多类别处理功能。以下代码即可将 pandas.DataFrame 转换为预测结果：","Kumo Tabular 依据 OpenMDW 许可协议 1.1 版本发布：/blog/nvidia/(https://openmdw.ai/license/1-1/)。NVIDIA 认为值得信赖的 AI 是共同责任，我们已建立政策和实践以支持广泛的 AI 应用开发。遵循服务条款下载或使用时，开发者应与其所支持的模型团队协作，以确保该模型满足相关行业和使用案例的要求，并防范不可预见的产品滥用。请在此报告模型质量、风险、安全漏洞或 NVIDIA AI 相关问题：https://github.com/NVIDIA/structured-data-models/issues。","我们感谢 David Holzmüller：https://dholzmueller.github.io/ 对 Kumo Tabular 提供的重要想法和消融实验。我们感谢 Vignesh Kothapalli：https://kvignesh1420.github.io/ 在实习期间对 Kumo Tabular 的帮助。"]},"en":{"title":"NVIDIA releases the open-source tabular foundation model Kumo Tabular, ranking first on four benchmarks including TabArena.","summary":"NVIDIA released Kumo Tabular, an open-source tabular foundation model that performs classification and regression on labeled tables in a single forward pass, without training, hyperparameter tuning, or feature engineering.","category":"Industry","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"NVIDIA releases the open-source tabular foundation model Kumo Tabular, ranking first on four benchmarks including TabArena. - Aioga AI News","description":"NVIDIA released Kumo Tabular, an open-source tabular foundation model that performs classification and regression on labeled tables in a single forward pass, without training, hype...","url":"https://www.aioga.com/en/news/tcf79upi169og7ftecq28pps2/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"8645150863444cf0","translatedAt":"2026-09-29T18:08:53.186Z"},"ja":{"title":"NVIDIA、オープンソースの表形式基盤モデル「Kumo Tabular」を公開、TabArenaなど4つのベンチマークで首位","summary":"NVIDIAは、オープンソースの表形式基盤モデル「Kumo Tabular」を公開しました。ラベル付き表形式データに対して単一のフォワード推論を行うだけで分類と回帰を実行でき、学習、ハイパーパラメータ調整、特徴量エンジニアリングは不要です。","category":"業界動向","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"NVIDIA、オープンソースの表形式基盤モデル「Kumo Tabular」を公開、TabArenaなど4つのベンチマークで首位 - Aioga AIニュース","description":"NVIDIAは、オープンソースの表形式基盤モデル「Kumo Tabular」を公開しました。ラベル付き表形式データに対して単一のフォワード推論を行うだけで分類と回帰を実行でき、学習、ハイパーパラメータ調整、特徴量エンジニアリングは不要です。","url":"https://www.aioga.com/ja/news/tcf79upi169og7ftecq28pps2/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"8645150863444cf0","translatedAt":"2026-09-29T18:09:16.104Z"},"ko":{"title":"NVIDIA, 오픈소스 테이블형 파운데이션 모델 Kumo Tabular 공개…TabArena 등 4개 벤치마크에서 1위","summary":"NVIDIA가 오픈소스 테이블형 파운데이션 모델 Kumo Tabular를 공개했습니다. 레이블이 있는 표에 대해 단 한 번의 순방향 추론만으로 분류와 회귀를 수행할 수 있으며, 학습이나 하이퍼파라미터 조정, 특성 엔지니어링이 필요하지 않습니다.","category":"업계 동향","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"NVIDIA, 오픈소스 테이블형 파운데이션 모델 Kumo Tabular 공개…TabArena 등 4개 벤치마크에서 1위 - Aioga AI 뉴스","description":"NVIDIA가 오픈소스 테이블형 파운데이션 모델 Kumo Tabular를 공개했습니다. 레이블이 있는 표에 대해 단 한 번의 순방향 추론만으로 분류와 회귀를 수행할 수 있으며, 학습이나 하이퍼파라미터 조정, 특성 엔지니어링이 필요하지 않습니다.","url":"https://www.aioga.com/ko/news/tcf79upi169og7ftecq28pps2/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"8645150863444cf0","translatedAt":"2026-09-29T18:08:36.872Z"},"es":{"title":"NVIDIA lanzó el modelo fundacional tabular de código abierto Kumo Tabular, que ocupa el primer puesto en cuatro benchmarks, incluido TabArena.","summary":"NVIDIA ha lanzado Kumo Tabular, un modelo fundacional de código abierto para datos tabulares, que permite realizar clasificación y regresión en tablas etiquetadas mediante una sola inferencia hacia adelante, sin necesidad de entrenamiento, ajuste de hiperparámetros ni ingeniería de características.","category":"Industria","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"NVIDIA lanzó el modelo fundacional tabular de código abierto Kumo Tabular, que ocupa el primer puesto en cuatro benchmarks, incluido TabArena. - Aioga Noticias de IA","description":"NVIDIA ha lanzado Kumo Tabular, un modelo fundacional de código abierto para datos tabulares, que permite realizar clasificación y regresión en tablas etiquetadas mediante una sola...","url":"https://www.aioga.com/es/news/tcf79upi169og7ftecq28pps2/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"8645150863444cf0","translatedAt":"2026-09-29T18:14:14.054Z"},"fr":{"title":"NVIDIA publie Kumo Tabular, un modèle fondamental open source pour les données tabulaires, classé premier dans quatre benchmarks, dont TabArena.","summary":"NVIDIA a publié le modèle de base tabulaire open source Kumo Tabular, qui permet de réaliser la classification et la régression sur des tableaux étiquetés avec une seule inférence avant, sans besoin d'entraînement, d'ajustement de paramètres ou d'ingénierie des fonctionnalités.","category":"Industrie","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"NVIDIA publie Kumo Tabular, un modèle fondamental open source pour les données tabulaires, classé premier dans quatre benchmarks, dont TabArena. - Aioga Actualités IA","description":"NVIDIA a publié le modèle de base tabulaire open source Kumo Tabular, qui permet de réaliser la classification et la régression sur des tableaux étiquetés avec une seule inférence...","url":"https://www.aioga.com/fr/news/tcf79upi169og7ftecq28pps2/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"8645150863444cf0","translatedAt":"2026-09-29T18:13:44.273Z"},"de":{"title":"NVIDIA veröffentlicht das Open-Source-Tabellen-Grundlagenmodell Kumo Tabular, das bei vier Benchmarks, darunter TabArena, den ersten Platz belegt.","summary":"NVIDIA hat das Open-Source-Tabellen-Basismodell Kumo Tabular veröffentlicht, das Klassifikation und Regression für tabellarische Daten mit Labels durch einen einzigen Vorwärtsdurchlauf durchführen kann, ohne Training, Feineinstellung oder Feature-Engineering.","category":"行业动态","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"NVIDIA veröffentlicht das Open-Source-Tabellen-Grundlagenmodell Kumo Tabular, das bei vier Benchmarks, darunter TabArena, den ersten Platz belegt. - Aioga KI-News","description":"NVIDIA hat das Open-Source-Tabellen-Basismodell Kumo Tabular veröffentlicht, das Klassifikation und Regression für tabellarische Daten mit Labels durch einen einzigen Vorwärtsdurch...","url":"https://www.aioga.com/de/news/tcf79upi169og7ftecq28pps2/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"8645150863444cf0","translatedAt":"2026-09-29T18:14:22.721Z"},"pt-BR":{"title":"A NVIDIA lançou o modelo básico de tabelas de código aberto Kumo Tabular, ocupando o primeiro lugar em quatro benchmarks, incluindo o TabArena","summary":"A NVIDIA lançou o modelo de base para dados tabulares de código aberto Kumo Tabular, que realiza classificação e regressão em tabelas rotuladas com uma única inferência de passagem direta, sem necessidade de treinamento, ajuste de hiperparâmetros ou engenharia de características.","category":"行业动态","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"A NVIDIA lançou o modelo básico de tabelas de código aberto Kumo Tabular, ocupando o primeiro lugar em quatro benchmarks, incluindo o TabArena - Aioga Notícias de IA","description":"A NVIDIA lançou o modelo de base para dados tabulares de código aberto Kumo Tabular, que realiza classificação e regressão em tabelas rotuladas com uma única inferência de passagem...","url":"https://www.aioga.com/pt-BR/news/tcf79upi169og7ftecq28pps2/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"8645150863444cf0","translatedAt":"2026-09-29T18:19:01.244Z"},"ru":{"title":"NVIDIA выпустила открытое табличное базовое моделирование Kumo Tabular, занявшее первое место в четырех бенчмарках, включая TabArena","summary":"NVIDIA выпустила открытую базовую табличную модель Kumo Tabular, которая выполняет классификацию и регрессию за один прямой проход по размеченным табличным данным — без обучения, настройки гиперпараметров или конструирования признаков.","category":"行业动态","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"NVIDIA выпустила открытое табличное базовое моделирование Kumo Tabular, занявшее первое место в четырех бенчмарках, включая TabArena - Aioga Новости ИИ","description":"NVIDIA выпустила открытую базовую табличную модель Kumo Tabular, которая выполняет классификацию и регрессию за один прямой проход по размеченным табличным данным — без обучения, н...","url":"https://www.aioga.com/ru/news/tcf79upi169og7ftecq28pps2/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"8645150863444cf0","translatedAt":"2026-09-29T18:18:53.517Z"},"ar":{"title":"أعلنت NVIDIA عن نموذج الأساس مفتوح المصدر للبيانات الجدولية Kumo Tabular، الذي احتل المرتبة الأولى في أربعة معايير مرجعية، من بينها TabArena.","summary":"أعلنت NVIDIA عن إطلاق نموذج Kumo Tabular الأساسي مفتوح المصدر للبيانات الجدولية، الذي يتيح إجراء التصنيف والانحدار من خلال استدلال أمامي واحد على الجداول ذات التسميات، من دون الحاجة إلى التدريب أو ضبط المعلمات أو هندسة الميزات.","category":"行业动态","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"أعلنت NVIDIA عن نموذج الأساس مفتوح المصدر للبيانات الجدولية Kumo Tabular، الذي احتل المرتبة الأولى في أربعة معايير مرجعية، من بينها TabArena. - Aioga أخبار الذكاء الاصطناعي","description":"أعلنت NVIDIA عن إطلاق نموذج Kumo Tabular الأساسي مفتوح المصدر للبيانات الجدولية، الذي يتيح إجراء التصنيف والانحدار من خلال استدلال أمامي واحد على الجداول ذات التسميات، من دون الحاج...","url":"https://www.aioga.com/ar/news/tcf79upi169og7ftecq28pps2/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"8645150863444cf0","translatedAt":"2026-09-29T18:19:24.542Z"},"hi":{"title":"एनवीआईडीआईए ने ओपन-सोर्स टेबल मॉडल Kumo Tabular जारी किया, जो TabArena सहित चार मानकों में पहले स्थान पर है","summary":"NVIDIA ने ओपन-सोर्स टेबलर फ़ाउंडेशन मॉडल Kumo Tabular जारी किया है। यह लेबल वाले टेबलर डेटा पर एक बार फ़ॉरवर्ड इन्फ़रेंस करके वर्गीकरण और रिग्रेशन कर सकता है; इसके लिए ट्रेनिंग, हाइपरपैरामीटर ट्यूनिंग या फ़ीचर इंजीनियरिंग की आवश्यकता नहीं है।","category":"行业动态","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"एनवीआईडीआईए ने ओपन-सोर्स टेबल मॉडल Kumo Tabular जारी किया, जो TabArena सहित चार मानकों में पहले स्थान पर है - Aioga AI समाचार","description":"NVIDIA ने ओपन-सोर्स टेबलर फ़ाउंडेशन मॉडल Kumo Tabular जारी किया है। यह लेबल वाले टेबलर डेटा पर एक बार फ़ॉरवर्ड इन्फ़रेंस करके वर्गीकरण और रिग्रेशन कर सकता है; इसके लिए ट्रेनिंग, हा...","url":"https://www.aioga.com/hi/news/tcf79upi169og7ftecq28pps2/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"8645150863444cf0","translatedAt":"2026-09-29T18:24:00.929Z"},"it":{"title":"NVIDIA ha rilasciato il modello fondazionale tabulare open source Kumo Tabular, classificatosi al primo posto in quattro benchmark, tra cui TabArena.","summary":"NVIDIA ha rilasciato il modello di base open source per dati tabellari Kumo Tabular, che consente di completare attività di classificazione e regressione su tabelle con etichette tramite una singola inferenza forward, senza necessità di addestramento, ottimizzazione dei parametri o feature engineering.","category":"行业动态","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"NVIDIA ha rilasciato il modello fondazionale tabulare open source Kumo Tabular, classificatosi al primo posto in quattro benchmark, tra cui TabArena. - Aioga Notizie IA","description":"NVIDIA ha rilasciato il modello di base open source per dati tabellari Kumo Tabular, che consente di completare attività di classificazione e regressione su tabelle con etichette t...","url":"https://www.aioga.com/it/news/tcf79upi169og7ftecq28pps2/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"8645150863444cf0","translatedAt":"2026-09-29T18:24:15.311Z"},"nl":{"title":"NVIDIA brengt het open-sourcetabelbasismodel Kumo Tabular uit en staat op de eerste plaats in vier benchmarks, waaronder TabArena","summary":"NVIDIA heeft het open-source tabelbasismodel Kumo Tabular uitgebracht, waarmee classificatie en regressie op gelabelde tabellen kunnen worden uitgevoerd met slechts één enkele voorwaartse inferentie, zonder training, afstemming of feature-engineering.","category":"行业动态","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"NVIDIA brengt het open-sourcetabelbasismodel Kumo Tabular uit en staat op de eerste plaats in vier benchmarks, waaronder TabArena - Aioga AI-nieuws","description":"NVIDIA heeft het open-source tabelbasismodel Kumo Tabular uitgebracht, waarmee classificatie en regressie op gelabelde tabellen kunnen worden uitgevoerd met slechts één enkele voor...","url":"https://www.aioga.com/nl/news/tcf79upi169og7ftecq28pps2/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"8645150863444cf0","translatedAt":"2026-09-29T18:24:03.917Z"},"tr":{"title":"NVIDIA, açık kaynaklı tablo biçimli temel model Kumo Tabular'ı yayımladı; TabArena dahil dört karşılaştırmalı testte birinci sırada yer aldı.","summary":"NVIDIA, etiketli tablolar üzerinde tek bir ileri geçiş çıkarımıyla sınıflandırma ve regresyon gerçekleştirebilen, eğitim, hiperparametre ayarı veya özellik mühendisliği gerektirmeyen açık kaynaklı tablo temel modeli Kumo Tabular'ı yayımladı.","category":"行业动态","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"NVIDIA, açık kaynaklı tablo biçimli temel model Kumo Tabular'ı yayımladı; TabArena dahil dört karşılaştırmalı testte birinci sırada yer aldı. - Aioga AI Haberleri","description":"NVIDIA, etiketli tablolar üzerinde tek bir ileri geçiş çıkarımıyla sınıflandırma ve regresyon gerçekleştirebilen, eğitim, hiperparametre ayarı veya özellik mühendisliği gerektirmey...","url":"https://www.aioga.com/tr/news/tcf79upi169og7ftecq28pps2/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"8645150863444cf0","translatedAt":"2026-09-29T18:28:36.882Z"},"vi":{"title":"NVIDIA phát hành mô hình nền tảng bảng dữ liệu mã nguồn mở Kumo Tabular, đứng đầu trong bốn bộ tiêu chuẩn đánh giá, bao gồm TabArena.","summary":"NVIDIA phát hành mô hình nền tảng bảng dữ liệu nguồn mở Kumo Tabular, có thể hoàn thành việc phân loại và hồi quy trên các bảng dữ liệu có nhãn chỉ bằng một lần suy luận truyền xuôi, không cần huấn luyện, tinh chỉnh tham số hay kỹ thuật đặc trưng.","category":"行业动态","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"NVIDIA phát hành mô hình nền tảng bảng dữ liệu mã nguồn mở Kumo Tabular, đứng đầu trong bốn bộ tiêu chuẩn đánh giá, bao gồm TabArena. - Tin tức AI Aioga","description":"NVIDIA phát hành mô hình nền tảng bảng dữ liệu nguồn mở Kumo Tabular, có thể hoàn thành việc phân loại và hồi quy trên các bảng dữ liệu có nhãn chỉ bằng một lần suy luận truyền xuô...","url":"https://www.aioga.com/vi/news/tcf79upi169og7ftecq28pps2/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"8645150863444cf0","translatedAt":"2026-09-29T18:28:26.599Z"},"id":{"title":"NVIDIA merilis model fondasi tabel sumber terbuka Kumo Tabular, yang menduduki peringkat pertama dalam empat tolok ukur, termasuk TabArena.","summary":"NVIDIA merilis model dasar tabel open-source Kumo Tabular. Model ini dapat melakukan klasifikasi dan regresi pada tabel berlabel hanya dengan satu kali inferensi forward pass, tanpa perlu pelatihan, penyetelan parameter, atau rekayasa fitur.","category":"行业动态","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"NVIDIA merilis model fondasi tabel sumber terbuka Kumo Tabular, yang menduduki peringkat pertama dalam empat tolok ukur, termasuk TabArena. - Berita AI Aioga","description":"NVIDIA merilis model dasar tabel open-source Kumo Tabular. Model ini dapat melakukan klasifikasi dan regresi pada tabel berlabel hanya dengan satu kali inferensi forward pass, tanp...","url":"https://www.aioga.com/id/news/tcf79upi169og7ftecq28pps2/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"8645150863444cf0","translatedAt":"2026-09-29T18:28:56.243Z"},"th":{"title":"NVIDIA เปิดตัวโมเดลพื้นฐานสำหรับข้อมูลแบบตารางแบบโอเพนซอร์ส Kumo Tabular ซึ่งครองอันดับหนึ่งในเกณฑ์มาตรฐาน 4 รายการ รวมถึง TabArena","summary":"NVIDIA เปิดตัวโมเดลพื้นฐานสำหรับตารางแบบโอเพนซอร์ส Kumo Tabular ซึ่งสามารถทำการจำแนกประเภทและการถดถอยบนข้อมูลตารางที่มีป้ายกำกับได้ด้วยการอนุมานครั้งเดียวแบบไปข้างหน้า โดยไม่จำเป็นต้องฝึกโมเดล ปรับจูนพารามิเตอร์ หรือทำวิศวกรรมคุณลักษณะ","category":"行业动态","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"NVIDIA เปิดตัวโมเดลพื้นฐานสำหรับข้อมูลแบบตารางแบบโอเพนซอร์ส Kumo Tabular ซึ่งครองอันดับหนึ่งในเกณฑ์มาตรฐาน 4 รายการ รวมถึง TabArena - ข่าว AI Aioga","description":"NVIDIA เปิดตัวโมเดลพื้นฐานสำหรับตารางแบบโอเพนซอร์ส Kumo Tabular ซึ่งสามารถทำการจำแนกประเภทและการถดถอยบนข้อมูลตารางที่มีป้ายกำกับได้ด้วยการอนุมานครั้งเดียวแบบไปข้างหน้า โดยไม่จำเป็น...","url":"https://www.aioga.com/th/news/tcf79upi169og7ftecq28pps2/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"8645150863444cf0","translatedAt":"2026-09-29T18:33:46.251Z"},"pl":{"title":"NVIDIA opublikowało otwartoźródłowy model tabelaryczny Kumo Tabular, zajmując pierwsze miejsce w czterech benchmarkach, w tym TabArena","summary":"NVIDIA wydała open-source’owy podstawowy model tabelaryczny Kumo Tabular, który umożliwia klasyfikację i regresję na oznaczonych tabelach za pomocą pojedynczego przejścia w przód, bez potrzeby trenowania, dostrajania parametrów ani inżynierii cech.","category":"行业动态","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"NVIDIA opublikowało otwartoźródłowy model tabelaryczny Kumo Tabular, zajmując pierwsze miejsce w czterech benchmarkach, w tym TabArena - Aioga Wiadomości AI","description":"NVIDIA wydała open-source’owy podstawowy model tabelaryczny Kumo Tabular, który umożliwia klasyfikację i regresję na oznaczonych tabelach za pomocą pojedynczego przejścia w przód,...","url":"https://www.aioga.com/pl/news/tcf79upi169og7ftecq28pps2/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"8645150863444cf0","translatedAt":"2026-09-29T18:33:07.479Z"}},"evidenceTier":"verified-news","reviewStatus":"editorial-selected","indexable":true,"editorialCover":""}}