AI Training Datasets

Dataset

privacy_guides_tw

Codeglacier

門羅幣 (Monero, XMR)： 8AEyT2biCC7S5impodPReHdRrUipEnj3GLFRZStyeUHiWWzq9q8emkpibJiPRxScXranJqKrinEere938k9pWkVcFWYTDzG 比特幣 (Bitcoin, BTC)： bc1qr0h3h9dlhzdu067psskwfsa8c0mfzaryv5626g

Procedurally generated indoor scenes with ground-truth path-traced renders and precomputed 3-slat VAE latents, built for training [RenderFormer]-style neural renderers. Each sample is one scene observed from 14 camera poses along an orbit. This is a [WebDataset]. Each sample carries: The latents hold three roles sharing one coordinate grid per frame (a unified-proxy Frames are concatenated along the token axis with a frameoffsets index rather than stored separately, so frame i is coords[frameoffsets[i]:frameoffsets[i+1]]. Token count varies slightly per frame (~2750) because light proxies are voxelized in per frame. Fields arrive as raw bytes. Decode them with: coords are integer…

Ranked match telemetry from the public Brawl Stars API, one row per ranked set. A set is up to three games on one fixed map and mode, first team to two wins; record holds the game-by-game outcome. Collected and published automatically by an open-source ETL pipeline. This page shows season54, the season in progress. It is refreshed every time the crawler runs, so the numbers above move during the season. Each season is its own config, because they are not interchangeable — see Limitations. Earlier seasons: season53. Each season is one file, written in time order. Parquet stores the min and max of every row group, so a reader asking for a date range skips the groups that cannot match rather…

最后更新 (last updated): 2026-10-17 11:32 CST (UTC+8) This is an unofficial mirror of the input data for the CICE sea ice model, as listed on the official CICE Input Data wiki page. All data are produced and published by the CICE Consortium on Zenodo. Files are the original, unmodified.tar.gz archives (not unpacked); MD5 checksums were verified against Zenodo. Not mirrored: the gx1 and tx1 JRA55 multi-year tar files (Zenodo 3731014, 10044769); they contain the same data as the per-year files included here. Mirrors the CICEdata/ grid organisation. Unpack each archive inside your CICEdata/ directory (tar xzf ), as described in the wiki. - gx3/ – 3° displaced-pole grid: grid + initial conditions…

Steered text and image generations from UniAR (ShareLab-SII/UniAR-RL), one of the unified vision-language models in the unified-vlm-steering project. Steering adds alpha vhat (the per-layer unit difference-of-means vector) to the residual stream at every layer of a layer config. Qwen3-VL backbone, 36 decoder layers (0-indexed). Images are BSQ tokens rendered by an SD3 decoder (16 decoding steps, 544 px). Every quadrant covers 7 concepts x 7 layer configs x 11 alphas x 20 prompts. Layer configs (0-indexed): L9 9, L18 18, L27 27, early 7-10, mid 16-19, late 25-28, all 0-35. - Quadrant names are 2: txt2txt and img2txt steer text generation, img2img and txt2img steer image generation. - alpha =…

Versioned runtime, task mods, and replay/evaluation records for AgentArk. Consumers should read manifest.json and registry/.jsonl instead of scanning directories. This dataset and the AgentArk runtime, task, and record artifacts in this repository are released under the Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0), unless otherwise noted. The materials may be used, copied, modified, and redistributed for non-commercial academic research and educational purposes with appropriate attribution. Commercial use, including use in commercial products, services, hosted platforms, paid evaluations, or proprietary training/evaluation pipelines, is not permitted…

Steered text and image generations from Emu3.5 (BAAI), one of the unified vision-language models in the unified-vlm-steering project. Steering adds alpha vhat (the per-layer unit difference-of-means vector) to the residual stream at every layer of a layer config. 34B dense autoregressive model, 64 decoder layers (0-indexed). Images are 32x32 IBQ tokens (512 px), generated on BAAI's patched vLLM engine with request-id-keyed CFG. The mid window in both sweeps is mid8 = layers 28-35 (it replaces the 4-layer mid4 window). vectors/steertablevllm.pt holds the per-layer unit vectors and layer masks the image engine used. Every quadrant covers 7 concepts x 7 layer configs x 11 alphas x 20 prompts.…

Dataset

tucson-mayor-council-meetings

Josh Estrada

Structured, machine-readable catalog of City of Tucson Mayor & Council meeting recordings, built from the city's own permanent mirror on the Internet Archive (collection cotaz), plus the caption files preserved with those uploads, plus speaker-labeled transcripts produced by diarization. The City of Tucson uploads its full Mayor & Council meeting recordings to YouTube and mirrors them to the Internet Archive under the cotaz collection - so the recordings are already preserved. But they are scattered across 75+ archive.org items with inconsistent titles, no unified dates, and no machine-readable catalog. This dataset is that catalog, and it also harvests the caption files (.en.vtt…

A set of prerender 2D images from GSO datasets, including png, mask, camera calibration and poses. When training Deep Learning models for Novel View Synthesis (NVS) or 3D-to-2D Representation Learning, loading 3D meshes and rendering views on-the-fly inside PyTorch DataLoaders creates massive bottlenecks. This script solves critical problems

Training rollouts of one verl run, one parquet per trainer step (data/step-.parquet), rows in the order the trainer consumed them; the rollout.n samples of a prompt are contiguous and share uid. prompttokenids / responsetokenids are the exact token ids (special tokens kept, no padding) and rolloutlogprobs the sampling-time log-prob of every response token. input / output are decoded with skipspecialtokens=True and are for reading only. score is the raw reward.

Rendered eye images provided by hyunsubs. Upload is in progress; the files currently present are a partial release. - data/ -.tar: image-only, uncompressed TAR archives, each approximately 5 GB or smaller. A source folder's final archive can be smaller. - manifests/ -.jsonl: one record per archived image, with its relative path, bytes, and sha256 checksum. Image paths inside each TAR preserve their original numbered source folder and file name, for example 01gray/leftclose1FinalImage.0000.jpeg. Folders ending in gray contain grayscale processing outputs. Existing source ZIP archives, model weights, training checkpoints, logs, caches, and experiment directories are excluded from this…

Dataset · Text to video

PAWBench-Results

YuandongPu

This dataset package defines the clean HF repository layout and contains the public metadata needed to run the PAWBench Full31 coverage and Table1 PAW-Cal PAWEval rerun inputs. In the local no-copy handoff, videos are not duplicated under this directory; they are uploaded into sharded videos/bymediaid/ subdirectories from the private upload plan after explicit human approval. - videos/bymediaid/ /.mp4: one physical video per mediaid after human upload from the private upload plan. - manifests/sharedmediaregistry.jsonl: public media identity registry. - manifests/full31pawevalmanifest.jsonl: Full31 logical PAWEval rows. - manifests/table1pawevalrerunrows.jsonl: Table1 PAW-Cal rerun rows.…