jingang

[David Holzmüller @DHolzmueller]

Super excited that TabICLv2 is out 🎉

🚀Beats RealTabPFN-2.5 with no tuning and purely synthetic pre-training data.

👉Introduces QASSMax for long-context generalization, early target embedding, repeated feature grouping, Muon, etc., and a much diversified synthetic data prior.

[Gael Varoquaux 🦋 @GaelVaroquaux]

🎉 Announcing TabICLv2: State-of-the art Table Foundation Model, fast and open source

A breakthrough for tabular ML: better prediction and faster runtime than alternatives, work by @JingangQu, @DHolzmueller, @MarineLeMorvan, and myself 👇

[David Holzmüller @DHolzmueller]

🚨Visit Jingang at our ICML poster later today if you are interested in
- scaling tabular foundation models to larger datasets
- open-source tabular foundation models
- meta-learning + in-context learning

[jingang @JingangQu]

@roydanroy Thanks !!!

[David Holzmüller @DHolzmueller]

TabICL offers two checkpoints:
v1: Version in the paper, open-source pre-training code
v1.1: Better, preview of upcoming work, same architecture, benchmarked in TabArena. Pre-training code not yet public. Outperforms TabPFNv2 for, roughly, >= 7,000 samples.
Stay tuned for v2 😀

[jingang @JingangQu]

@yuvalbenj Thanks !

[jingang @JingangQu]

@roydanroy @GaelVaroquaux Hello Dan, I just tweeted my poster info at https://t.co/SRpuUrhr5A. Looking forward to seeing you !

[jingang @JingangQu]

I’ll present "TabICL: A Tabular Foundation Model for In‑Context Learning on Large Data" at ICML 2025.

🗓 Tuesday, July 15, 2025 | 4:30–7:00 PM PDT
📍 East Exhibition Hall A‑B, Booth hashtag#E‑320

If you're curious about TabICL, come by the poster — I'd love to chat !

[jingang @JingangQu]

I’ll present "TabICL: A Tabular Foundation Model for In‑Context Learning on Large Data" at ICML 2025.

🗓 Tuesday, July 15, 2025 | 4:30–7:00 PM PDT
📍 East Exhibition Hall A‑B, Booth hashtag#E‑320

If you're curious about TabICL, come by the poster — I'd love to chat !

[Gael Varoquaux 🦋 @GaelVaroquaux]

👨‍🎓🧾✨#icml2025 Paper: TabICL, A Tabular Foundation Model for In-Context Learning on Large Data
With @JingangQu, @DHolzmueller and @MarineLeMorvan

TL;DR: a well-designed architecture and pretraining gives best tabular learner, and more scalable
1/9

[jingang @JingangQu]

@frecciarossa_fr C'est parfait. Quand le train arrive-t-il à Lyon Part Dieu ? Il n'y a aucune information sur ce train dans la salle d'attente.