aab
Datasets
All datasets matching “aab”AA-Briefcase-Lite
AA-Briefcase-Lite
The public example scenario for AA-Briefcase, Artificial Analysis' frontier agentic evaluation of realistic, long-horizon knowledge work.
Leaderboard and detailed results
Launch article
AA-Briefcase extends frontier model benchmarking beyond coding and short-form reasoning to the professional deliverables knowledge workers produce day to day. It consists of four private scenarios in which agents complete realistic professional workflows across data science… See the full description on the dataset page: https://huggingface.co/datasets/ArtificialAnalysis/AA-Briefcase-Lite.fasterwam
fasterwam — Franka Panda teleoperated demonstrations
Four tasks, 100 episodes each. Every task folder has the same flat layout:
<task>/episode_0.hdf5 … <task>/episode_99.hdf5
<task>/merge_manifest.json maps each episode to the original recording it came from
Task
Description
Recorded
Details
place_cube/
Place a cube in a bowl
2026-09-10
place_cube/RECROP.json — crops re-derived 2026-09-18 from the stored raw frames with the same window as the other three tasks… See the full description on the dataset page: https://huggingface.co/datasets/aabyaneh/fasterwam.StateBench-v1
StateBench v1
110 million verified state episodes for pretraining, SFT, and on-policy SDFT
Train models to retain, edit, address, and transform state across long execution traces.
Choose a configuration
Configuration
Scale
Use when
default
10,000,000 programs · 7.80 GiB
You need 36 explicit families and varied surfaces
dense-100m
100M episodes · 98.61 GiB
You need high-throughput state training with grounded, structurally varied… See the full description on the dataset page: https://huggingface.co/datasets/aabbdev/StateBench-v1.cawm
cawm
Tactile manipulation data for world-model training: real-robot teleoperation on
a Franka Panda arm, plus matched simulation rollouts. All streams are 10 Hz.
Preview
Each clip is wrist camera · agentview camera (where one was recorded) ·
FlexiTac contact map, played at 9.6× real time — the data is 10 Hz, so
these are every 8th frame at 12 fps.
These are excerpts, not whole trajectories: a contiguous window of a single
episode, chosen automatically as the… See the full description on the dataset page: https://huggingface.co/datasets/aabyaneh/cawm.simready-assets-pr-test5simready-assets-pr-test2