Joined August 2022
- Tweets 682
- Following 198
- Followers 1,098
- Likes 2,862
178 Photos and videos
daisugi: v0.0.7
continuing to expand the supported tree machines & also registered a few to tidymodels.
1
2
91
announcing a new project called daisugi 🌳
daisugi is a new support interface for not-so-well-known tree machines
daisugi (v0.0.3.9000) currently supports:
- Yggdasil
- WildWood
- NGBoost
- Perpetual
- SnapBoost
- EBM
1
2
4
326
introducing KimiTab 🌊
a Tabular Foundation Model built with Kimi K2.6 Agent!
KimiTab architecture is based on TabICL, Orion, TabPFN, & TabDPT, with two-stage synthetic & real data in mind.
The repo open sources pre-training, training, distillation, docs methodology.
1
5
361
179
Ambassador Frank Hull ☤ retweeted
May 26
After seeing these tweets, I decided to try it out on my own old Ubuntu computer with RTX 1070 GPU (the one that I just upgraded from 16.04 all the way to 24.04 the other day). Asked Codex on my Mac to connect to that machine and install and test qwen3 8b. So far really impressive - running 30 tok/s!
ok this is wild. 10 year old gtx 1080 8gb pascal card running qwen3 8b locally at 18-20 tok/s via hermes agent and it's actually doing the thing.
asked it to build a wireworld cellular automata simulator with 10 tests. autonomous run, no hand holding. expected it to fail on the tool calls. that's not what happened.
write_file works. browser_navigate works. terminal commands work. file ops, package installs, version probes, environment setup. agent is firing tool calls cleanly and the model is reasoning about next steps at 18-20 tok/s. on hardware that pre dates "agentic" as a word.
it even hit an npm install fail because node 12 is too old. didn't crash. didn't ask me. just started bootstrapping nvm on its own to fix the environment.
10 minutes in. 40% context used. 7.5gb of 8gb vram occupied. still going.
i did not think this would work on this hardware. this is the most i've been wrong this month.
18
30
309
79,108
Ambassador Frank Hull ☤ retweeted
are we even ready for TabPFN-3-Plus?
1
5
185
Ambassador Frank Hull ☤ retweeted
May 12
TabPFN-3 is live. A massive leap forward in scale, speed & accuracy. 1M data points and 10-1000x faster inference on one H100. No training. No tuning. Load your dataset and predict.
#tabpfn #tabularfoundationmodels #priorlabs
1
6
29
1,699
Ambassador Frank Hull ☤ retweeted
May 11
Today, we are excited to announce the first release of Synthefy-Tabular: our new class of tabular foundation models.
While Migas is focused on time-series forecasting, Synthefy-Tabular is built for a broader class of prediction problems: estimating demand, revenue, risk, prices, costs, conversion, churn, and other real-world numerical outcomes from structured data.
We are open-sourcing the full stack behind the model, including the synthetic data generator, training recipes, inference configs, and evaluation setup.
Across a ~100-dataset regression benchmark spanning TabArena, TALENT, and OpenML, Synthefy-Tabular achieves state-of-the-art aggregate results in our evaluation, outperforming TabPFN-2.6 while being smaller and faster at inference.
One of the most exciting parts of this release is thinking mode: an inference-time option that lets the model optimize its context before making predictions.
With Synthefy-Tabular, our goal is to make Synthefy the best foundation-model platform for real-world prediction — not just forecasting. Full model code and technical report coming soon.
5
9
2,484
huge news in the tabular world
with a focus on 'non-neural-network methods'
🌲🌳🌴
May 6
Today I want to share one of the main projects I have been working on: TabulAI. Tabular data runs much of the business world, yet it has not received the same sustained research attention as images, text, audio, or code. TabulAI exists to change that.
1/9
1
126
friendly reminder, you can now try out TabICLv2 from your phone 🤳🏻
1
95
a huge update to Prior Labs's ⚡awesome-tabpfn⚡ repo via the ambassadors!
listing 200 papers & repos showcasing use cases of Tabular Foundation Models 🔥
check out the link shared below ⬇️
1
3
154
Ambassador Frank Hull ☤ retweeted
Delighted to announce my new paper, "Training-Free Probabilistic Time-Series Forecasting with Conformal Seasonal Pools". 🔥🔥
The result: a training-free conformal model obliterates Amazon's DeepNPTS on every metric, by orders of magnitude in significance.
3
7
13
1,000
Ambassador Frank Hull ☤ retweeted
May 4
Huge news: @prior_labs has signed a definitive agreement to be acquired by @SAP.
€1B invested over four years to build a globally-leading frontier AI lab for structured data — in Europe, in the open.
Independent entity. Same team, same mission, same open models. A massive boost to what we can do. The mission just got accelerated.
Founders’ statement: priorlabs.ai/blog-posts/prio…
(Deal subject to regulatory approval; terms not disclosed.)
36
32
512
51,747
TabICLv2 is now available to use from your browser 🎉
frankiethull-tabicl-studio.s…
It ships with well known datasets but you can also upload your own csv
🎉 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 👇
4
12
2,038