datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
atari-breakout-dataset
Atari-Breakout Dataset
This is a large dataset of 10M video frames and actions collected from the Breakout atari environment (Bellemare et al., 2012) in order to train world models.The dataset enables reproducible, large-scale experiments in action-conditioned video prediction. It is meant to be used with Jasmine, our JAX-based world modeling codebase.
Dataset Summary
Environment: Atari Learning Environment
Frames: 10 million
Resolution: 84 × 84
Format:… See the full description on the dataset page: https://huggingface.co/datasets/p-doom/atari-breakout-dataset.doom-dataset
Doom Dataset
This is a large dataset of 10M video frames and actions collected from the Doom environment for training world models.The dataset enables reproducible, large-scale experiments in action-conditioned video prediction. It is meant to be used with Jasmine, our JAX-based world modeling codebase.
Dataset Summary
Environment: VizDoom
Frames: 10 million
Resolution: 60 × 80
Format: ArrayRecord (for fast I/O)
Splits: train / val / test
License: CC0 1.0… See the full description on the dataset page: https://huggingface.co/datasets/p-doom/doom-dataset.doom-dataset-largedenemedatabaseOptimized binary datasets for Project X.
Status: Confidential / Work in Progress
Since this is a confidential project, the files are encrypted. Please do not delete the files. Thank you :)
doom-rnd-largecrowd-code-dataset-1.0
Install crowd-code 2.0 to help crowd-source the next-generation coding dataset.
crowd-code-dataset-1.0 is an anonymized dataset of fine-grained IDE interactions crowd-sourced across 25 people over the last 6 months using crowd-code 1.0, a VS Code/Cursor extension capturing large parts of the software engineering workflow.
The dataset captures real research engineering workflows (character-level edits, navigation, terminal use, iterative… See the full description on the dataset page: https://huggingface.co/datasets/p-doom/crowd-code-dataset-1.0.common-craft
common-craft
common-craft is a dataset of Minecraft survival gameplay videos curated for research on world modeling.
It can be combined with scalable repositories such as Jasmine.
Overview
Total duration: ~5,000 hours
Content type: Survival-mode Minecraft Let's Plays
Source: YouTube videos and playlists (handpicked)
Format: Raw videos (.mp4 and .webm) at a resolution of 640x360 and 30 FPS along with full metadata
Structure
common-craft/
├──… See the full description on the dataset page: https://huggingface.co/datasets/p-doom/common-craft.doom-dataset-1atari-crazy_climber-dataset
Atari-Crazy Climber Dataset
This is a large dataset of 10M video frames and actions collected from the Crazy Climber atari environment (Bellemare et al., 2012) in order to train world models.The dataset enables reproducible, large-scale experiments in action-conditioned video prediction. It is meant to be used with Jasmine, our JAX-based world modeling codebase.
Dataset Summary
Environment: Atari Learning Environment
Frames: 10 million
Resolution: 84 × 84… See the full description on the dataset page: https://huggingface.co/datasets/p-doom/atari-crazy_climber-dataset.doom-dense-arnold-latents
DoomDiT per-tic latents
Derived from the public raw dataset RohanNaga/doom-dense-arnold: every tic of each episode encoded once through a frozen Stable Diffusion autoencoder, stored as one .npy per episode plus a .npz sidecar with the per-tic metadata. Released so that a training run can start without re-encoding, and so that latents are reproducible bit for bit: latent bytes depend on the autoencoder, its scaling, the precision and the encode batch size, all recorded in… See the full description on the dataset page: https://huggingface.co/datasets/RohanNaga/doom-dense-arnold-latents.atari-assault-dataset
Atari-Assault Dataset
This is a large dataset of 10M video frames and actions collected from the Assault atari environment (Bellemare et al., 2012) in order to train world models.The dataset enables reproducible, large-scale experiments in action-conditioned video prediction. It is meant to be used with Jasmine, our JAX-based world modeling codebase.
Dataset Summary
Environment: Atari Learning Environment
Frames: 10 million
Resolution: 84 × 84
Format:… See the full description on the dataset page: https://huggingface.co/datasets/p-doom/atari-assault-dataset.atari-battle_zone-dataset
Atari-Battle Zone Dataset
This is a large dataset of 10M video frames and actions collected from the Battle Zone atari environment (Bellemare et al., 2012) in order to train world models.The dataset enables reproducible, large-scale experiments in action-conditioned video prediction. It is meant to be used with Jasmine, our JAX-based world modeling codebase.
Dataset Summary
Environment: Atari Learning Environment
Frames: 10 million
Resolution: 84 × 84
Format:… See the full description on the dataset page: https://huggingface.co/datasets/p-doom/atari-battle_zone-dataset.atari-amidar-dataset
Atari-Amidar Dataset
This is a large dataset of 10M video frames and actions collected from the Amidar atari environment (Bellemare et al., 2012) in order to train world models.The dataset enables reproducible, large-scale experiments in action-conditioned video prediction. It is meant to be used with Jasmine, our JAX-based world modeling codebase.
Dataset Summary
Environment: Atari Learning Environment
Frames: 10 million
Resolution: 84 × 84
Format:… See the full description on the dataset page: https://huggingface.co/datasets/p-doom/atari-amidar-dataset.atari-boxing-dataset
Atari-Boxing Dataset
This is a large dataset of 10M video frames and actions collected from the Boxing atari environment (Bellemare et al., 2012) in order to train world models.The dataset enables reproducible, large-scale experiments in action-conditioned video prediction. It is meant to be used with Jasmine, our JAX-based world modeling codebase.
Dataset Summary
Environment: Atari Learning Environment
Frames: 10 million
Resolution: 84 × 84
Format:… See the full description on the dataset page: https://huggingface.co/datasets/p-doom/atari-boxing-dataset.atari-alien-dataset
Atari-Alien Dataset
This is a large dataset of 10M video frames and actions collected from the Alien atari environment (Bellemare et al., 2012) in order to train world models.The dataset enables reproducible, large-scale experiments in action-conditioned video prediction. It is meant to be used with Jasmine, our JAX-based world modeling codebase.
Dataset Summary
Environment: Atari Learning Environment
Frames: 10 million
Resolution: 84 × 84
Format: ArrayRecord… See the full description on the dataset page: https://huggingface.co/datasets/p-doom/atari-alien-dataset.atari-bank_heist-dataset
Atari-Bank Heist Dataset
This is a large dataset of 10M video frames and actions collected from the Bank Heist atari environment (Bellemare et al., 2012) in order to train world models.The dataset enables reproducible, large-scale experiments in action-conditioned video prediction. It is meant to be used with Jasmine, our JAX-based world modeling codebase.
Dataset Summary
Environment: Atari Learning Environment
Frames: 10 million
Resolution: 84 × 84
Format:… See the full description on the dataset page: https://huggingface.co/datasets/p-doom/atari-bank_heist-dataset.atari-asterix-dataset
Atari-Asterix Dataset
This is a large dataset of 10M video frames and actions collected from the Alien atari environment (Bellemare et al., 2012) in order to train world models.The dataset enables reproducible, large-scale experiments in action-conditioned video prediction. It is meant to be used with Jasmine, our JAX-based world modeling codebase.
Dataset Summary
Environment: Atari Learning Environment
Frames: 10 million
Resolution: 84 × 84
Format:… See the full description on the dataset page: https://huggingface.co/datasets/p-doom/atari-asterix-dataset.coinrun-dataset
CoinRun Dataset
This is a large dataset of 50M video frames and actions collected from the CoinRun environment (Cobbe et al., 2020) for training world models.The dataset enables reproducible, large-scale experiments in action-conditioned video prediction. It is meant to be used with Jasmine, our JAX-based world modeling codebase.
Dataset Summary
Environment: CoinRun (Procgen Benchmark)
Frames: 50 million
Resolution: 64 × 64
Format: ArrayRecord (for fast I/O)… See the full description on the dataset page: https://huggingface.co/datasets/p-doom/coinrun-dataset.RADAR_datasetdoomaiidm-eval
IDM Eval Set
A validation set for evaluating Inverse Dynamics Models on macOS screen recordings. Each sample is a 5-second clip of real productivity desktop usage (browser, IDE, terminal, docs, dashboards) paired with a ground-truth action log captured at the OS level.
The task: given a short screen recording, predict the sequence of user input actions (keypresses, mouse clicks, scrolls, cursor moves) that produced the observed screen changes.
Code
Training… See the full description on the dataset page: https://huggingface.co/datasets/p-doom/idm-eval.openai-minecraft-datasetThis dataset is a preprocessed version of the OpenAI Minecraft Video Dataset VPT, reformatted into array record files compatible with the Grain dataloader for efficient streaming.
AGI-CAST-0.6k
AGI-CAST: Behaviour-Cloning Knowledge Work
AGI-CAST-0.6k is a >600-hour dataset of screencasts capturing raw workflows of researchers at p(doom). This is the first major release in our effort to collect months-long fine-grained expert trajectories of human reasoning in their day-to-day work.
Please refer to the blog post for more details.
Dataset Summary
Contemporary frontier model development has saturated the internet and… See the full description on the dataset page: https://huggingface.co/datasets/p-doom/AGI-CAST-0.6k.crowd-code-dataset-0.1The crowd-code-dataset-0.1 is a raw, unfiltered dataset of fine-grained IDE interactions collected during the development of Jasmine using crowd-code, a VS Code/Cursor extension capturing large parts of the software engineering workflow.
The dataset captures real research engineering workflows (character-level edits, navigation, terminal use, iterative debugging). The crowd-code-dataset-0.1 only includes data from the Jasmine authors. We are actively working on cleaning and curating the full… See the full description on the dataset page: https://huggingface.co/datasets/p-doom/crowd-code-dataset-0.1.atari-demon_attack-dataset
Atari-Demon Attack Dataset
This is a large dataset of 10M video frames and actions collected from the Demon Attack atari environment (Bellemare et al., 2012) in order to train world models.The dataset enables reproducible, large-scale experiments in action-conditioned video prediction. It is meant to be used with Jasmine, our JAX-based world modeling codebase.
Dataset Summary
Environment: Atari Learning Environment
Frames: 10 million
Resolution: 84 × 84… See the full description on the dataset page: https://huggingface.co/datasets/p-doom/atari-demon_attack-dataset.doom-e1-internet-gameplayI've used the Inverse Dynamic Model, I've previously trained on manually recorded gameplay, on pure gameplay YouTube videos.
This dataset is in public domain, use it however you want.
doom-dense-arnold
DoomDiT dense Arnold recordings
Lossless, per-tic recordings of the Arnold agent (Lample and Chaplot, AAAI 2017) playing ViZDoom deathmatch
with 8 bots on Freedoom assets, made for training action-conditioned world models. Every engine tic
(35 per second) is stored with the executed control vector, so the data can be used at any frame stride.
Recorded September 2026 at CMU for the DoomDiT project (Rohan Nagabhirava, Keerthana Chirumamilla).
What is here… See the full description on the dataset page: https://huggingface.co/datasets/RohanNaga/doom-dense-arnold.doomalay-superpowers
doomalay-superpowers
The obra/superpowers agent-skills
corpus, hub-native for the doomalay public library (v3): every skill is
a whole-directory BUNDLE (SKILL.md + scripts + references + prompts riding
one {"v":1,"entry":"SKILL.md","files":[…]} manifest), the docs ride as
hidden doc items inside the superpowers-obra bunch, and the maintainer
shell scripts ride the script library. items/index.json is the item list;
each item's payload lives at its file path.
What's… See the full description on the dataset page: https://huggingface.co/datasets/ScoobyBaby1999/doomalay-superpowers.DOOMGAN-Ocular-Morphs
DOOMGAN: Ocular Morph Dataset
This repository contains the official public dataset for the paper: "DOOMGAN: High-Fidelity Dynamic Identity Obfuscation Ocular Generative Morphing" funded by the NSF award no. 2345561.
The dataset consists of 10,000 high-fidelity morphed ocular images generated by the DOOMGAN model. These images are intended to facilitate research and development of Morph Attack Detection (MAD) systems for visible-spectrum ocular biometrics.
Paper: IJCB 2025 DOOMGAN… See the full description on the dataset page: https://huggingface.co/datasets/BharathK333/DOOMGAN-Ocular-Morphs.doom-dungeon-55
doom-dungeon-55
Checkpoint archive for the doom_v55_hg world-model lineage. The canonical, fully documented serving copy of the robot-campaign checkpoint (v6sf_phase2_ep26/) lives at alakazamworld/doom-dungeon-hg; start there.