AI Training Datasets

Observations meteo brutes regrillees, mises a jour en continu. Sources: NNJA-AI (Brightband/NOAA, CC-BY 4.0), EUMETSAT Data Store, NOAA/NCEI, ISD/HadISD.

Gated internal mirror. Each dataset lives under a neutral code (data-NNN/). Clip shards follow the p4d-1 / p4d-1.2 WebDataset format (one tar member per frame and modality,.idx.json member offsets for range reads), with a per-dataset data-NNN/meta.tar holding dataset.json (splits, licenses/provenance, conventions, depthsemantics, quality, multiview), index.parquet (one row per clip) and the shard indices. The exception is data-013, which holds per-(sequence, camera) source units read by its own streaming loader. multiplies the per-pixel depth regression weight (lossweights(status, confidence=...)); see dataset.json modalities.depthconfidence.rule / calibration. resolution as depth) on real…

Dataset

atlas-32-turn-level-actor-critic

Peijia Qin

Read this first. The reading copy of this directory is t2ance/atlas-experiments under 32-turn-level-actor-critic/; the saved training steps and the per-token training arrays are only in the Hugging Face repository t2ance/atlas-32-turn-level-actor-critic. Does a critic that predicts the return at the start of each turn, and is supervised there alone, learn on the orchestrator's rollouts (report 30's token-level critic did not), and does a policy trained on its turn-level advantages keep reading candidates where reading pays (report 31's critic-free training dropped to 0.93 explores a problem with the accuracy of its submissions unchanged)? - W&B runs, in pqin/atlas-grpo: the smoke…

Research snapshot of official legislation collected from Consultoria Juridica / consultoria.gov.do. Not legal advice. Official gazettes / government portals prevail over this corpus. - data/laws.parquet — one row per instrument (id, title, full text, dates, status, identifiers, article count, source URL). - data/articles.parquet — one row per article/section (lawid, article number, title, text, source URL). - scrapers/collectdo.py — research collector used to build this snapshot. - scrapers/common.py — shared collector helpers. - scrapers/worldlib.py — shared fetch (live official URL, Wayback fallback). - scrapers/archivefallbacks.py — Wayback / Common Crawl CDX helpers. Official government…

This repository is the immutable audit copy of raw Weibo API responses collected by a private archive. Original gzip snapshots are stored unchanged inside bounded TAR shards with SHA-256 manifests. Older snapshots remain individually addressable. The database records each repository, entry and byte range. The parsed and authoritative text lives in the operator's PostgreSQL database; these files exist for audit and re-processing without re-fetching from Weibo. This dataset uses Hugging Face manual gated access. Access requests are reviewed by the repository owner. Do not mirror, redistribute, bulk-download, or use the responses to identify or contact individuals. Source content may contain…

Dataset · Other

stubhub-dataset

Rebrowser.net

prettyname: "StubHub Ticket Marketplace Dataset" - open-data - data-science - web-scraping - scraper - crawler - data-collection - stubhub - tickets - event-tickets - secondary-market - resale-tickets - live-events - sports-tickets - concert-tickets sizecategories: taskcategories: - other - configname: event-listings datafiles: - configname: events datafiles: - configname: venues datafiles: Daily snapshots of StubHub resale ticket listings, events, and venues with seating details, delivery types, and availability data across sports, concerts, and theater. This dataset is a preview sample of the StubHub dataset published by Rebrowser. If you're doing academic research, you may be eligible…

Dataset

albedo

Dendrite Holdings

Reference trajectories from SN97 (albedo) evaluations. Each row is a - sampleid — source shard and row of the underlying task ( /.parquet:: ) - messages — list: a system prompt, the user task (PR description + instructions), then alternating assistant reference steps and user environment observations - run — which of the sample's reference runs this row is (1-3) - madeedit — whether the run changed any file (null for rows recorded before 2026-09-22) - score — the run's score on the sample's final checklist, 0-1, on the same scale miners are scored on (null before 2026-09-22) - questions — JSON list of {id, text, milestone, rung, weight, earned}: the checklist the sample was scored on and…

RVCBench Accepted to NeurIPS 2026. RVCBench is a benchmark dataset for studying robustness in voice cloning, text-to-speech, speaker privacy, audio protection, adversarial audio perturbations, and related audio generation pipelines. RVCBench is designed for evaluating how modern voice cloning (VC), TTS, and audio

This repository preserves Nemotron-CC D004 inputs, data and resume indexes, native training checkpoints, and validation outputs. Each checkpoint and validation directory retains its own run or attempt identity so outputs from different weights are not combined. This archive section maps six seed-42 GGLR ablations and nine seed-42/43/44 evaluations for GGLR, Sign Original, and Sign Normalized. The listed D004 validations use the step-23,564 weights and each records 100,638,720 shifted targets. Validation directory names encode 4gpu-b48 where that suffix is present. GGLR seed 42 is the rerun-07 run, and the centroid-driven-weight-update-only ablation is rerun-02. These identities are kept…