datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
doom-dataset-largedoom-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.doom-dataset-1crowd-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.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.SauerkrautLM-Doom-MultiVec-31k
SauerkrautLM-Doom-MultiVec-31k
31,645 human gameplay demonstration frames for training the SauerkrautLM-Doom-MultiVec-1.3M DOOM action classifier.
This dataset was recorded by a human player in VizDoom's SPECTATOR mode across 4 recording sessions totaling approximately 2 hours of gameplay in the defend_the_center scenario. Each frame includes the ASCII game view, real VizDoom depth buffer data, and soft action labels derived from keyboard input.
Dataset Structure… See the full description on the dataset page: https://huggingface.co/datasets/VAGOsolutions/SauerkrautLM-Doom-MultiVec-31k.doom-e1-gameplayThis is some crappy gameplay of me playing DOOM (1993) first episode ("Knee Deep in Hell") with VizDoom on a resolution of 320x240 to train an Inverse Dynamic Model on DOOM.
Use it how you want. The gameplay is in the public domain.
AGI-CAST-idm-actions
AGI-CAST IDM Actions
We release model-generated keyboard, mouse-click, cursor-movement and scroll annotations for the AGI-CAST-0.6k screencast dataset.
The labels were generated with p-doom/idm, an inverse dynamics model trained to recover user input actions from short windows of screen recordings.
Please refer to the AGI-CAST blog post and the IDM blog post for more context.
Code
These labels were generated with p-doom/idm.
The… See the full description on the dataset page: https://huggingface.co/datasets/p-doom/AGI-CAST-idm-actions.doomalay-metrics
doomalay-metrics
Doomalay hub library data.
doomalay-themes
doomalay-themes
Doomalay hub library data.
forge-gpt5-doom-mips-3step
forge-gpt5-doom-mips-3step
Harbor evaluation on terminal-bench@2.0: 0/1 resolved (0.0%), 0 errors
Dataset Info
Rows: 1
Columns: 25
Columns
Column
Type
Description
instance_id
Value('string')
Task identifier (e.g. astropy__astropy-12907)
reward
Value('float64')
Verifier reward (e.g. 0.0 or 1.0)
resolved
Value('bool')
Whether the task was resolved (reward > 0)
agent
Value('string')
Agent name used for this trial
model
Value('string')
Model… See the full description on the dataset page: https://huggingface.co/datasets/reasoning-degeneration-dev/forge-gpt5-doom-mips-3step.forge-gpt5-doom-mips-full
forge-gpt5-doom-mips-full
Harbor evaluation on terminal-bench@2.0: 0/1 resolved (0.0%), 0 errors
Dataset Info
Rows: 1
Columns: 25
Columns
Column
Type
Description
instance_id
Value('string')
Task identifier (e.g. astropy__astropy-12907)
reward
Value('float64')
Verifier reward (e.g. 0.0 or 1.0)
resolved
Value('bool')
Whether the task was resolved (reward > 0)
agent
Value('string')
Agent name used for this trial
model
Value('string')
Model… See the full description on the dataset page: https://huggingface.co/datasets/reasoning-degeneration-dev/forge-gpt5-doom-mips-full.CV-Mistral-Hackathon_doom-mistral-final-latest-10-frames-ShareGPTOriginal set is one frame per sample.
This does the latest 10 samples, like so:
Sample 1: Frames 1, 2, 3, 4, 5, 6, 7, 8, 9, 10
Sample 2: Frames 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
Sample 3: Frames 3, 4, 5, 6, 7, 8, 9, 10, 11, 12
Sample 4: Frames 4, 5, 6, 7, 8, 9, 10, 11, 12, 13
If it's trained this way maybe it will have a better understanding of the ASCII since it's not doing 0-shot every time, and gets some previous information.
training_datamf-doomThis dataset is designed to generate lyrics with HuggingArtists.doom-defend-the-center-100k-oracle
Doom Defend the Center 100K oracle
100K rows dataset from 317 episodes of Doom Defend the Center scenario.
A programmatic oracle plays the scenario with access to privileged ViZDoom info: spin one way, freeze and fire when an enemy crosses the crosshair.
The dataset does not contain traces of privileged info and can be learned from RGB only.
Each row carries
frames.u8: one 100x160 RGB frame at time t.
labels.npz: turn (left/none/right) and shoot (yes/no) + actions at time t-2… See the full description on the dataset page: https://huggingface.co/datasets/anakin87/doom-defend-the-center-100k-oracle.forge-gpt54-doom-mips-full
forge-gpt54-doom-mips-full
Harbor evaluation on terminal-bench@2.0: 0/1 resolved (0.0%), 0 errors
Dataset Info
Rows: 1
Columns: 25
Columns
Column
Type
Description
instance_id
Value('string')
Task identifier (e.g. astropy__astropy-12907)
reward
Value('float64')
Verifier reward (e.g. 0.0 or 1.0)
resolved
Value('bool')
Whether the task was resolved (reward > 0)
agent
Value('string')
Agent name used for this trial
model
Value('string')
Model… See the full description on the dataset page: https://huggingface.co/datasets/reasoning-degeneration-dev/forge-gpt54-doom-mips-full.DoomFrameDataset
DoomFrameDataset
DoomFrameDataset is a ViZDoom frame-action dataset generated from policy rollouts. It is packaged as WebDataset tar shards for streaming training, imitation learning, behavior cloning, and offline reinforcement-learning experiments.
The dataset contains RGB game frames paired with the action selected by the rollout policy and per-step metadata such as reward, episode id, step id, terminal flag, and value estimate.
Dataset Size
Config
Files… See the full description on the dataset page: https://huggingface.co/datasets/brahmandam/DoomFrameDataset.Doom-source-codedoom-mistral-finaldoom-mixtral-textUniHypoBenchdoom-mixtral-alpacalw_questions_from_claude_more_doomodokodoko-chan-datasetdoom_mathdoom_physicsdoom-actions-gemini-bot