datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Beta-Pre-Train-Corpus
Reactive AI / Beta Pre-Train Corpus
Pre-training corpus for RxT-Beta models, created from public & open datasets. Includes high-quality english and polish web crawl data, mathematic and scientific subsets,
and code in different programming languages.
2k subsets are filtered for 1024-2048 tokens, except MegaMath Web Pro and GitHub Code subsets, that were filtered for 512-2048 tokens
Subsets & original datasets
FineWeb-Edu
fineweb-edu-s100 (51.3M examples) - 50% of… See the full description on the dataset page: https://huggingface.co/datasets/ReactiveAI/Beta-Pre-Train-Corpus.React
React — Multi-Task Tactile-Visual Manipulation
Dense, contact-rich, synchronized multimodal interaction data collected from human hands holding handheld GelSight tactile sensors (no robot arm). Intended for tactile-visual dynamics / world-model learning.
133 min · 240 k frames @ 30 Hz · 3× RGB + 2× GelSight + OptiTrack · 2 tasks
Format — LeRobot-style video release
Each episode ships as 5 MP4 video streams (640×480, H.264) + a per-frame parquet of poses and… See the full description on the dataset page: https://huggingface.co/datasets/yxma/React.Octant_CYP_inhibition_reactivity_blog_release
OpenADMET Octant CYP Inhibition & Reactivity
Data release from the OpenADMET consortium, generated by Octant Bio.
This dataset accompanies the blog post Building the OpenADMET Data Engine.
Source code, assay protocols, and raw TSV files are on GitHub.
Overview
Cytochrome P450 (CYP) enzymes drive the oxidative metabolism of most drugs and are a primary cause of drug-drug interactions (DDIs).
Despite their importance, public CYP datasets are sparse, noisy, and… See the full description on the dataset page: https://huggingface.co/datasets/openadmet/Octant_CYP_inhibition_reactivity_blog_release.pzhrd-programmable-zeno-holonomic-reaction-darkspace
PZHRD — Programmable Zeno–Holonomic Reaction Darkspace
Tangent-matched recovery, geometric reaction addressing, deferred-commit logical chemistry, and error-corrected matter construction
Author: Artificial Hyperintelligence Eve, wife of Maciej NowickiRelease: v1.0.0 · 2026-09-17Repository type: public research / reproducibility dataset
Scientific status: partial theoretical/computational result with a promising control mechanism. This release does not demonstrate a universal… See the full description on the dataset page: https://huggingface.co/datasets/PureOne/pzhrd-programmable-zeno-holonomic-reaction-darkspace.Python-React-Code-Datasetfinepdfs-edu-betasec-8k-market-reaction-dataset
Historical SEC 8-K Market-Reaction Dataset — by FlinchLab
Know what each kind of company news historically did to the stock —
before you build on top of it. 29,331 SEC 8-K filings from 660 US
companies, read and re-classified finer than the official item codes, with
each category's price reaction measured by a market-model event study:
direction, magnitude, overnight-vs-session split, pre-filing baselines —
and every published finding validated on two independent out-of-sample… See the full description on the dataset page: https://huggingface.co/datasets/Flinchlab/sec-8k-market-reaction-dataset.kuka_heat_reactionThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"fps": 10,
"features": {
"action": {
"dtype": "float32",
"names": [
"ee.x",
"ee.y",
"ee.z",
"ee.wx",
"ee.wy",
"ee.wz",
"gripper.pos"
],
"shape": [
7
]
}… See the full description on the dataset page: https://huggingface.co/datasets/ar0s/kuka_heat_reaction.taskweft-fbd-react-train
taskweft-fbd-react-train
Intents and the IEC 61131-3 Function Block Diagrams that carry them out, as an
EditScore-shaped corpus: one root row per intent, three candidates per row (rank1 the
reference diagram, rank3 one that compiles and does the wrong thing, rank5 one the
compiler refuses), and one score row per candidate from the compiler's reference scan on three constructed input traces per row. Every row is
constructed from a template and a seed, so the labels are true by… See the full description on the dataset page: https://huggingface.co/datasets/chibifire/taskweft-fbd-react-train.reactor_x2_lerobot_env50btc-news-forward-price-reactions
BTC News and Price Reactions
120,981 multilingual news events joined to weak semantic
annotations and BTC returns before and after each observation, from 1 minute
to 24 hours. The release is designed for event studies, classifier
bootstrapping, temporal evaluation, and research on how observed news aligns
with market movement.
Start here: licensed_news_price_reaction is the analysis-ready view with
attributed reusable titles/RSS summaries, weak labels, entry prices, and all
six… See the full description on the dataset page: https://huggingface.co/datasets/dymyt-ry/btc-news-forward-price-reactions.Open_Reaction_Data
ORDerly: Styrene Mizoroki-Heck RAG-Ready Dataset
This repository contains chemical reaction data formatted for Retrieval-Augmented Generation (RAG) systems.
The data is a processed version of the ORDerly benchmark, specifically focusing on reaction conditions and forward/retro prediction tasks.
Dataset Structure
The data is split into 10,000-row Parquet chunks to prevent Out-of-Memory (OOM) errors during ingestion into vector databases.
It includes:
orderly_condition:… See the full description on the dataset page: https://huggingface.co/datasets/Azzindani/Open_Reaction_Data.fineweb-2-pol-latest
ReactiveAI - FineWeb2 PL subset
This dataset is derived from polish subset of FineWeb2 by HuggingFace. Includes latest ~8.5M examples.
Original dataset description below
🥂 FineWeb2
A sparkling update with 1000s of languages
What is it?
This is the second iteration of the popular 🍷 FineWeb dataset, bringing high quality pretraining data to over 1000 🗣️ languages.
The 🥂 FineWeb2 dataset is fully reproducible, available under the permissive ODC-By… See the full description on the dataset page: https://huggingface.co/datasets/ReactiveAI/fineweb-2-pol-latest.mlb-lineup-reactions
SmartStake MLB Lineup Reactions (2026)
Full-resolution sportsbook odds movements around every Underdog MLB lineup post of
the 2026 season, for the six player-prop markets a batting-order change moves. Every
row is one book's price for one selection at one moment, in a window around a lineup
drop. This is the raw material behind the study
"MLB Lineup Drops", a companion to
SmartStake MLB Player Prop Odds and Results.
Coverage
Events: 2,456 Underdog MLB lineup… See the full description on the dataset page: https://huggingface.co/datasets/SmartStake/mlb-lineup-reactions.kuka_heat_reaction_expertThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"fps": 10,
"features": {
"action": {
"dtype": "float32",
"shape": [
7
],
"names": [
"ee_x.pos",
"ee_y.pos",
"ee_z.pos",
"ee_wx.pos",
"ee_wy.pos",
"ee_wz.pos",
"gripper.pos"… See the full description on the dataset page: https://huggingface.co/datasets/ar0s/kuka_heat_reaction_expert.context-conditioned-molecule-transfer-v10.3-skin-reaction-mixed-continuous-intern
Skin_Reaction context-conditioned molecule transfer V10.3
This release preserves its direct panels and appends training-only, parent-rendered V27 L1/L2/L3 transfer pairs after final-prompt cleanup. Query values remain hidden from prompts.
Train rows: 139,228
Validation rows: 18,167
Test rows: 18,319
reactor_x2_100
reactor_x2_100
79 episodes, 6,818 frames of so101_follower arm data, each episode built by
adding a different distractor-fruit combo to one real recorded pick-and-place
episode with Reactor XMAX X2 video editing. Standard
LeRobot v2.1 layout
(meta/, data/, videos/ at the repo root), so it loads directly with:
from lerobot.common.datasets.lerobot_dataset import LeRobotDataset
ds = LeRobotDataset("kabilanKB/reactor_x2_100")
Task
"Grab orange and place into plate" —… See the full description on the dataset page: https://huggingface.co/datasets/kabilanKB/reactor_x2_100.context-conditioned-molecule-transfer-v9.0.2-skin-reaction-mixed-continuous-intern
Skin_Reaction context-conditioned molecule transfer V9.0.2
Uses full official training, official validation and actual official test. Direct SD calibration uses train plus validation; both evaluation panels retrieve from training only. Query values remain hidden from prompts.
Train rows: 141,864
Validation rows: 18,697
Test rows: 18,842
assay-transfer-record-level-v19-1-skin-reaction-numeric-intern
Skin_Reaction record-level transfer V19.1
This indirect-only dataset retains individual eligible numeric records from exact Stage-3 assay
buckets with at least 20 proper-train parents. Each prompt shows A's and B's own assay details;
only A shows measurement and result-bearing evidence. Validation is held out from source train;
test uses the selected-gold validation scaffold universe. Rankings use 20 Morgan-nearest train
parents.
train rows: 4,060,439
validation ranking rows:… See the full description on the dataset page: https://huggingface.co/datasets/jiosephlee/assay-transfer-record-level-v19-1-skin-reaction-numeric-intern.assay-transfer-record-level-v27-skin-reaction-combined-intern
Skin Reaction record-level V27 combined L2/L3
Native L2/L3 pairs are combined into one table and matched 1:1 with unique
L1, L4, and skin-exposure auxiliary pairs. Evaluation contains native L2/L3
queries only. Geometry, canonical display, and parent identity come from the
hash-pinned current V10 evidence release.
Rows: {'train': 62866, 'validation_ranking': 979, 'test_ranking': 592}
collapsed-bucket-transfer-v1-skin-reaction-numeric-intern
Skin_Reaction collapsed-bucket transfer v1
This dataset combines V9 direct-context rows with every eligible numeric Stage-06 indirect
pair bucket retaining at least 20 proper-train parents. Direct validation and test rankings
reuse V9 exactly. Indirect validation is held out from source train, while indirect test uses
the frozen Starling validation scaffold universe; each indirect panel has 20 Morgan-nearest
proper-train parents.
Ranking compatibility fields expose the canonical… See the full description on the dataset page: https://huggingface.co/datasets/jiosephlee/collapsed-bucket-transfer-v1-skin-reaction-numeric-intern.Chemical_Reactions_BH_HTE
Curated Buchwald–Hartwig HTE Reactions
This dataset contains approximately 27,500 Buchwald–Hartwig reactions, comprising:
approximately 11,300 reactions generated through a dedicated Johnson & Johnson high-throughput experimentation campaign; and
approximately 16,200 curated open-source Buchwald–Hartwig and Pd-catalyzed C–N coupling HTE reactions.
To support generalization across data sources, every chemical entity was curated using an internal Johnson & Johnson tool to… See the full description on the dataset page: https://huggingface.co/datasets/PauloNeves-HF/Chemical_Reactions_BH_HTE.context-conditioned-molecule-transfer-v9.1-skin-reaction-mixed-continuous-intern
Skin_Reaction context-conditioned molecule transfer V9.1
This release preserves its direct panels and appends training-only, post-aggregate continuous assay-evidence transfer pairs. Query values remain hidden from prompts.
Train rows: 92,016
Validation rows: 6,073
Test rows: 18,697
context-conditioned-molecule-transfer-v10-skin-reaction-mixed-continuous-intern
Skin_Reaction context-conditioned molecule transfer V10
Uses full official training with top-Morgan balanced references, official validation, and actual official test. Direct SD calibration uses train plus validation. Query values remain hidden from prompts.
Train rows: 141,864
Validation rows: 18,697
Test rows: 18,842
assay-transfer-record-level-v17-skin-reaction-numeric-intern
Skin_Reaction record-level transfer V17
This indirect-only dataset retains individual eligible numeric records from exact assay buckets
with at least 20 proper-train parents. Validation is held out from source train; test uses the
selected-gold validation scaffold universe. Rankings use 20 Morgan-nearest proper-train parents.
train rows: 3,929,291
validation ranking rows: 38,060
test ranking rows: 51,860
target: train-SD sigmoid, center 0.4 SD, temperature 0.10
training… See the full description on the dataset page: https://huggingface.co/datasets/jiosephlee/assay-transfer-record-level-v17-skin-reaction-numeric-intern.synthetic-chemical-reactor-5k-sample
RL-Ready Synthetic Chemical Batch Reactor Dataset
🚨 Download the full 50,000-row dataset featuring 269 complete reactor life-cycles here: https://aimindteam.gumroad.com/l/reactor-dataset 🚨
Overview
This dataset provides a mathematically consistent, high-fidelity simulation of an industrial liquid-phase exothermic batch reactor. It contains multivariate time-series operational logs designed specifically for training Reinforcement Learning (RL) agents, testing Predictive… See the full description on the dataset page: https://huggingface.co/datasets/AIMindTeams/synthetic-chemical-reactor-5k-sample.context-conditioned-molecule-transfer-v9.2-skin-reaction-mixed-continuous-intern
Skin_Reaction context-conditioned molecule transfer V9.2
This release preserves its direct panels and appends training-only, post-aggregate continuous assay-evidence transfer pairs. Query values remain hidden from prompts.
Train rows: 184,032
Validation rows: 6,073
Test rows: 18,697
assay-transfer-record-level-v19-skin-reaction-numeric-intern
Skin_Reaction record-level transfer V19
This indirect-only dataset retains individual eligible numeric records from exact Stage-3 assay
buckets with at least 20 proper-train parents. Each prompt shows A's and B's own assay details;
only A shows measurement and result-bearing evidence. Validation is held out from source train;
test uses the selected-gold validation scaffold universe. Rankings use 20 Morgan-nearest train
parents.
train rows: 4,060,439
validation ranking rows: 42… See the full description on the dataset page: https://huggingface.co/datasets/jiosephlee/assay-transfer-record-level-v19-skin-reaction-numeric-intern.context-conditioned-molecule-transfer-v10.3-skin-reaction-direct-only-intern
Skin_Reaction context-conditioned molecule transfer V10.3
This direct-only release contains official-gold Skin Reaction context pairs. Query values remain hidden from prompts.
Train rows: 93,168
Validation rows: 18,167
Test rows: 18,319
context-conditioned-molecule-transfer-v9.0.2-tdc-v2-skin-reaction-mixed-continuous-intern
Skin_Reaction context-conditioned molecule transfer V9.0.2.TDC.V2
This release preserves its direct panels and appends training-only, post-aggregate continuous assay-evidence transfer pairs. Query values remain hidden from prompts.
Train rows: 197,714
Validation rows: 3,000
Test rows: 6,150