AI Training Datasets

Derived from cvml-nus/assembly101 (Sener et al., "Assembly101: A Large-Scale Multi-View Video Dataset for Understanding Procedural Activities", CVPR 2022), licensed CC BY-NC 4.0 and gated by its authors. This derivative keeps that licence: non-commercial use only, with attribution to the original dataset and paper; please also accept the original dataset's terms. Access to this repository is gated likewise. Contents: per camera view, every stride-th decoded frame (stride = ceil(fps) // 4: 15 at 60 fps, i.e. 4 fps), short side resized to 256 px (aspect kept, antialiased bilinear), JPEG quality 90, in the TAS frame-cache "jpeg" format: /.s.ss256.jpg.bin (the JPEGs back to back) +.jpg.idx.npy…

AuraGym provides training datasets for security agentic coding tasks from open-source repositories. It was built via AuraForge, a synthetic data pipeline to build agentic training environments at scale. The synthetic variant contains 679 feature implementation tasks from 344 real-world repositories across Python, JavaScript, and TypeScript, covering 177 CWE categories. - AuraGymh: 431 tasks from 199 repositories, with human-written security tests. The variants overlap but neither is a subset of the other: 368 task IDs are shared, 63 occur only in the human variant, and 311 occur only in the synthetic variant. Together they contain 742 distinct task IDs across 1,110 records. All four…

Agent trajectory data for AI-for-Science tasks, collected across 8 frontier models (glm-5.2, claude-sonnet-5, opus-4.8, deepseek-v4-pro, qwen3.7-max, minimax-m3, gemini-3.5-flash, gpt-5-mini) under real agentic harnesses (Claude Code / Codex / Gemini CLI), each running in an isolated container with held-out data and a host-side evaluation service. Two subsets capture two distinct task shapes: 70 tasks built from published frontier papers, each framed as premise → tension → (what step to take). The agent must make an open-ended move (new regime / mechanism / method-correction …); rollouts are classified by discovery regime (search / retrieval / discovery / unverified) under Wang & Buehler…

This dataset was created using LeRobot.

Dataset · Object detection

CCOD_cotton_flower_detection

AgML

CCOD (Crop Canopy Organ-level Object Detection) is a UAV dataset of small plant organs seen from above crop canopies: cotton flowers and cotton bolls in Xinjiang, potato flowers at Keshan Farm in Heilongjiang, and rice panicles in Jiangsu, China. Images were captured at flowering or maturity from 12 m with the RGB camera of a DJI Mavic 3 Multispectral drone, cropped into 640 x 640 tiles (512 x 512 for rice), and annotated with dense bounding boxes, using a SAHI-YOLOv10n model to propose boxes that annotators then corrected. The cropcanopy column records which subset each image comes from, and the split column keeps the authors' 7:2:1 train/val/test split. The dataset contains 3,986 images…

Where should a set of transmitters be placed in a city? CityDeploy-Data pairs urban geometry with feasible transmitter sets and radio-simulation labels for scene-conditioned utility learning, spatial-set planning and is one deployment, not one TX. The primary corpus has 54,900 separate family memberships. The adaptation extension adds 3,590 unique deployments with 3,600 memberships and no archived maps. All 67 local scene packages are and five supplementary scene packages. Do not count supplementary geometry as additional primary benchmark training data. Read the complete scalar tables without downloading meshes or Radio Maps: Use streaming=True for row iteration, or select scale128…

This repository contains uncompressed tar shards for the fastwam dataset. The archive members keep their paths relative to the dataset root. Create a directory named fastwam and extract every file under that directory so the original relative paths are restored, for example: The public inventory is available as DATASETINVENTORY.json.gz, and the current shard state is in SHARDS.json. The inventory contains relative file paths and sizes only; it does not contain the source host path. This upload excludes latent caches and other excluded cache, output, and log directories (9 excluded directories are recorded in the inventory). It also excludes any other files omitted by the inventory.…

This release contains public-safe firing heat-load events for the cooperative ceramics maintenance program. Internal identifiers and non-public workshop equipment are excluded from the release.

Dataset

ContextBench

Context Language Models

ContextBench measures what an agent keeps, updates, and offloads when its input does not fit in its context window. Each task streams a sequence of operations as ordinary user turns, more than the context budget holds, and the reward is read from the The all config holds every instance. This release has 256 instances: 64 kvstore, 56 logtriage, 80 needleretention, 56 sudokusketchpad (32 levels, seeds 42-49), with a context budget of 32768 tokens and a reserve of 2048. Each operation is shown to the agent as === {header} ({remaining} op(s) remaining after this) === followed by its body. To evaluate an agent from Python, install the contextbench package, write an instance to disk, and pass…

Dataset · Feature extraction

Contextual-Embeddings

Brigham Young University - Idaho

We are releasing all 41,995 verses from the "Standard Works" (sourced from public-domain editions), each paired with an LLM-generated chapter-context block and a 3072-dimension embedding of the two combined. This dataset is an independent test of Anthropic's Contextual Retrieval technique on a domain its original evaluation did not cover: short, archaic, formulaic religious text from antiquity. The technique was introduced in Introducing Contextual Retrieval (Anthropic, 2024), which describes two components — Contextual Embeddings (prepend generated context before embedding) and Contextual BM25 (prepend it before lexical indexing). Anthropic reported that Contextual Embeddings alone reduced…

A local derivative of the D3IL demonstrations with two-camera vision coverage across all eight task subsets, LeRobot v2.1 trajectories, and precomputed LingBot-VA video/text features. Upload status: See UPLOADSTATUS.json. This repository is complete only when its status is complete. Files may appear incrementally during the initial upload. There are 25,195 train clips and 2,878 eval clips, split by episode. The release contains 8,298 Parquet trajectories, 16,596 MP4 videos, and 56,146 per-camera latent files. - lerobot/ /: meta/, data/, videos/, latents/, and per-subset configuration. - lerobot/lingbotjointconfig.json: shared action normalization and camera/channel mapping.…