Prior Labs Releases TabPFN-3.5: A Tabular Foundation Model That Beats the Winning Otto Kaggle Solution With Default Settings
https://www.marktechpost.com/2026/09/15/prior-labs-releases-tabpfn-3-5-a-tabular-foundation-model-that-beats-the-winning-otto-kaggle-solution-with-default-settings/📌 免調參打敗十年Kaggle冠軍：TabPFN-3.5的表格基礎模型野心
TL;DR：不用訓練、不用調參，TabPFN-3.5用預設設定贏過2015年Kaggle冠軍解法。
2015年那場Otto Group產品分類賽，3,505支隊伍搶10,000美元獎金，冠軍是兩位曾登上Kaggle grandmaster世界第一的高手，用36個模型疊出的複雜stacking方案。十一年後，一個從沒看過這份資料、只靠合成資料預訓練的模型，用預設設定、一分鐘內就把成績刷過了冠軍。
🤔 AutoGluon追了六年，最後一哩路最難
Otto Group Product Classification Challenge要求參賽者用93個經過模糊化處理的計數特徵，把商品分到9個類別，以multi-class log loss評分（越低越好）。冠軍Gilberto Titericz與Stanislav Semenov的方案是基於手工特徵的36模型多層stack。AutoGluon共同創辦人、現任Prior Labs研究員Nick Erickson多年來持續追蹤這個分數：AutoGluon在2020年論文中排名第23，2023年的1.0版衝到第14，2026年8月的1.6版來到第9。但最後一段路特別硬，從排名50到排名10，log loss只從0.41降到0.40；但從排名10要追到冠軍的0.382，還得再降0.018，幾乎是前段的兩倍努力。
🧩 一次前向傳播，沒有per-dataset訓練
Prior Labs推出的TabPFN-3.5是一款表格基礎模型（tabular foundation model），核心特色是直接對一張表做forward pass進行預測，不需要針對個別資料集做訓練或調參。根據Erickson的說法，TabPFN-3.5在Otto資料集上是用原始資料、預設設定跑出來的，在一臺RTX PRO 6000 GPU上僅花約一分鐘。這個模型只用合成資料預訓練，從未見過Otto或任何Kaggle資料集。相關的可重現Kaggle notebook已公開。
📊 私有排行榜0.375，七項benchmark第一
TabPFN-3.5在Otto私有排行榜上跑出0.375，優於冠軍的0.382。技術報告則指出它在TabArena、BeyondArena、STRABLE、MulTaBench、RelArena-α、TALENT、ScoringBench共七項benchmark拿下第一，不過拿冠軍的並非總是基礎模型本身：TabPFN-3.5-Thinking在TabArena、BeyondArena、STRABLE、MulTaBench四項稱王，而一個內部的TabPFN-Rel harness預覽版則拿下RelArena-α。
在涵蓋51個資料集的TabArena上，Thinking版Elo達1910，基礎模型1866，領先TabFM+的1823。Prior Labs表示基礎模型以130 Elo的差距、五分之一的時間打敗AutoGluon 1.6 extreme。在涵蓋142個資料集、包含分組、時間序列、寬表、文字密集與高基數資料的BeyondArena上，TabPFN-3.5比先前總排行榜第一名高出約150 Elo，但在分組、時間序列與大資料量的子集上，經過調參與ensemble的MLP仍然領先。
⚠️ KV cache沒漲，但大訓練集反而變慢
儘管參數量約為TabPFN-3的四倍，KV cache大小卻大致維持不變，快取後的單筆預測速度也與TabPFN-3相當；但在大訓練集情境下，基礎模型的運算速度最多會比TabPFN-3慢兩倍。此外，開放權重版本可在本機執行，適用於研究、評估與Kaggle，但正式生產環境需要透過Prior Labs的API或取得商業授權。
🎯 實務啟示
對於需要快速在表格資料上取得基準成績的工程師，TabPFN-3.5提供了一個「先跑一次forward pass」再決定是否需要客製化pipeline的新起點，尤其適合原型驗證與資料探索階段；但涉及大規模訓練資料或生產部署時，仍需評估授權條件與推論速度的實際權衡。
🔗 來源
- 標題：Prior Labs Releases TabPFN-3.5: A Tabular Foundation Model That Beats the Winning Otto Kaggle Solution With Default Settings
- 作者／機構：Asif Razzaq, MarkTechPost
- 連結：https://www.marktechpost.com/2026/09/15/prior-labs-releases-tabpfn-3-5-a-tabular-foundation-model-that-beats-the-winning-otto-kaggle-solution-with-default-settings/
#TabPFN #TabularML #FoundationModel #Kaggle #AutoML #MachineLearning #DataScience #AutoGluon #PriorLabs #Benchmark
原始資料 MarkTechPost · 收集於 2026-09-16
摘要原文
Prior Labs has released TabPFN-3.5 , the newest version of its tabular foundation model. It predicts on a table in a forward pass, with no per-dataset training or tuning. Prior Labs reports first place across 7 tabular benchmarks. A separate demonstration shows it beating the winning solution of a famous 2015 Kaggle competition. Deployable? Yes, with a license. Open weights run locally for research, evaluation and Kaggle, but production use needs Prior Labs’ API or a commercial license. The Otto Group Product Classification Challenge ran on Kaggle in 2015. It drew 3,505 teams competing for $10,000. Entrants sorted products into 9 categories using 93 obfuscated count features. Submissions were scored with multi-class log loss, where lower is better. The winning solution came from Gilberto Titericz and Stanislav Semenov. Both have held the world #1 Kaggle grandmaster ranking. Their entry was a multi-layer stack of 36 models built on hand-crafted features. Nick Erickson , co-creator of AutoGluon and an AI researcher at Prior Labs, has chased that score for years. According to Erickson, AutoGluon placed rank 23 in its 2020 paper . AutoGluon 1.0 reached rank 14 in 2023, and AutoGluon 1.6 reached rank 9 in August 2026. The final stretch was the hardest. Moving from rank 50 to rank 10 cut log loss from 0.41 to 0.40. Reaching the winning 0.382 from rank 10 took a further 0.018, nearly double. TabPFN-3.5 scores 0.375 on the private leaderboard. Erickson says it ran on raw data with default settings. It took about a minute on an RTX PRO 6000 GPU. The model was pretrained only on synthetic data and never saw Otto or any Kaggle dataset. A reproducible Kaggle notebook is public. The technical report lists first place on TabArena, BeyondArena, STRABLE, MulTaBench, RelArena-α, TALENT and ScoringBench. The top entry is not always the base model. TabPFN-3.5-Thinking holds first on TabArena, BeyondArena, STRABLE and MulTaBench. An internal TabPFN-Rel harness preview takes RelArena-α. On TabArena , a living benchmark of 51 datasets, Thinking reaches 1910 Elo. The base model scores 1866, ahead of TabFM+ at 1823. Prior Labs says the base model beats AutoGluon 1.6 extreme by 130 Elo in a fifth of the time. BeyondArena spans 142 datasets with grouped, temporal, wide, text-rich and high-cardinality data. TabPFN-3.5 finishes about 150 Elo points ahead of the previous overall leader. Tuned and ensembled MLPs still lead on its grouped, temporal and large-data slices. The KV cache stays roughly the same size as TabPFN-3, despite about 4x more parameters. Cached single-row predictions match TabPFN-3 speed. On large training sets, however, the base model runs up to 2x slower than TabPFN-3. Check out the Technical report and Technical details . All credit goes to the researcher of this project. Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter . Wait! are you on telegram? now you can join us on telegram as well. Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us The post Prior Labs Releases TabPFN-3.5: A Tabular Foundation Model That Beats the Winning Otto Kaggle Solution With Default Settings appeared first on MarkTechPost .
由 claude-code/sonnet 自動生成