AI Training Datasets

Dataset · Text retrieval

ipfs_korea_laws_ir

Justicedao

Research retrieval release of endomorphosis/ipfskorealaws (revision a174ce3d2e51385dd21584ba4aa7618c97910d86) packaged as country-laws-ir-graphrag/v1 (layout family skillcenter-huggingface-release/v3 / publicus-ir). Not legal advice. This is a research snapshot. The official gazette / authentic source of Korea (ROK) prevails over this corpus. Retrieved documents and graph edges are retrieval evidence only. No legal text was invented. Primary key: entrycid (CIDv1 raw sha2-256 of a canonical identity record). Integer documentindex values are compact shard pointers, not identities. Target Hub id (packaging metadata only): justicedao/ipfskorealawsir. entrycid, lawcid, recordtype, jurisdiction…

Municipal law of Brazil, harvested from SAPL (Sistema de Apoio ao Processo Legislativo, Interlegis / Senado Federal) instances run by camaras municipais. Part of the world municipal-law coverage programme (one repo per country, html.parquet per jurisdiction, shared worldmunicipal schema). (21.3%); this dataset is being filled tranche by tranche from that census. of 2025 only. (0 full-2025 harvests + 4 pilot seed files). Each SAPL instance exposes a Normas Juridicas search (/norma/pesquisar). Per law, the structured texto articulado (article-level HTML) is taken where the instance provides it; otherwise the official original-text PDF is extracted as one document-level row. Instance content…

Dataset

laos-transport-data

Boun Mee

Open data from Laos Bus Tickets: border checkpoints, bus, train and flight routes, the Laos–China Railway timetable and the Laos transport map. laosbustickets.com is the original source; this repository is a copy, synced every day by a GitHub Action. If the two ever differ, trust the website, and please cite its pages. - checkedAt: when the timetable was last read from the railway's app (UTC). If a daily check fails, the previous timetable is kept, so this date tells you how fresh it is. - dates: the days covered (the app sells tickets 7 days ahead). - trains[]: train (D88, C95, K11…), direction, from, to, runsOn (dates), weekdays, weekdaysNotOnSaleYet (days the app has not opened yet, not…

This repository defines the exact image sets used in PolypRegimeBench, a controlled benchmark of 12 polyp detectors across 6 training regimes. It contains manifests only: file lists, split assignments, filtered IDs, augmentation recipes and checksums. It does not redistribute any third-party endoscopy images, masks or videos. White-light images only. Narrow-band images were excluded so that modality is not confounded with the other factors. Split seed 42, fixed for all runs. Test sets are almost entirely polyp-positive (0, 5 and 0 negative images respectively). The augmentations transform image and mask together: flips, affine (scale, translation, rotation, shear), blur, colour/contrast and…

Dataset · Text generation

tb21-qwen3.8-27b-terminus2

OpenGuardrails

Raw, unedited agent trajectories for three full runs over the same 89 Terminal-Bench 2.1 tasks, plus every remediation re-run, so each number below can be checked against what the agents actually did. The gaps are not statistically significant. Paired over the same tasks: Qwen vs DeepSeek+terminus-2 is net +6 (exact McNemar p=0.29); Qwen vs DeepSeek+dsh is net +3 (p=0.68). At 89 tasks and one attempt each, a lead this size is what run-to-run noise produces. Read the rows as "comparable", not as a ranking. 1. Start from what each run actually scored. 2. Trials lost to infrastructure were re-run under the repaired environment; the re-run's verdict replaces the original. 3. A trial whose work…

Transcipts of @Axiomism youtube channel contents in various languages Axiomism youtube kanalinin video, shorts ve yayinlarinin cesitli dillerde transkriptleri Translations are made by the latest best AI available to the public, original language is Turkish https://www.youtube.com/watch?v=4DtePT40AVo Please add new videos/shorts/streams transcripts when new ones uploaded Can add new translations but please use best AI available to do so Created with the consent of @Axiomism

Dataset · Text retrieval

ipfs_belgium_laws_ir

Justicedao

Research retrieval release of endomorphosis/ipfsbelgiumlaws (revision ) packaged as country-laws-ir-graphrag/v1 (layout family skillcenter-huggingface-release/v3 / publicus-ir). Not legal advice. This is a research snapshot. The official gazette / authentic source of Belgium prevails over this corpus. Retrieved documents and graph edges are retrieval evidence only. No legal text was invented. Primary key: entrycid (CIDv1 raw sha2-256 of a canonical identity record). Integer documentindex values are compact shard pointers, not identities. Target Hub id (packaging metadata only): justicedao/ipfsbelgiumlawsir. entrycid, lawcid, recordtype, jurisdiction, language, instrumentid, instrumenttitle…

Opaque archive shards.

EgoDex Retarget — 16 robots EgoDex egocentric human demonstrations retargeted onto 16 robot embodiments. For every episode, the human wrist and hand motion is solved into a joint trajectory for each robot, and the robot is rendered and composited into the original scene, over hand-erased frames and per-frame metric depth. Each (robot, task) pair is one LeRobot v3.0 dataset. VR-VLA/VR-egodex-inpaint-cache-full (hand-erased frames), VR-VLA/VR-egodex-depth-cache-full

The data here seems good enough on its own to train a pretty strong NNUE. In my own tests it did a good job of pulling draws out of Stockfish:P To keep costs down, it was relabeled with t1-256x10-rl-base-swa-3860000, and most of the self-play games were also generated with that same model, in a single pass with a bit of temperature. Bigger lc0 models would be far more effective, but this one hit around 660k/s on 8× 5090s, which made it cheap, so that's what I went with.

Synthetic image-prompt pairs used as the reference distribution for MGFlow text-to-image post-training in The same reference images support both image-only and joint image-text matching. Model checkpoints and fitted reference distributions: The dataset contains 301,706 synthetic images at 512×512 with English prompts. All images were generated with the official distilled [FLUX.2 [klein] 4B](https://huggingface.co/black-forest-labs/FLUX.2-klein-4B) model using four sampling steps and guidance scale 1. Reference construction follows the procedure used by iRDM and AMFD: generate 24 candidates per caption, and retain the three with the highest PickScore. GenEval correctness test, considering up…

TRIAGE-Bench is a benchmark for evaluating how LLMs resolve conflicts between authoritative clinical knowledge sources. It covers 2,000 items across four conflict types, each with an explicit governing policy that defines policy-consistent correctness. - 2,000 benchmark items (500 per conflict type) grounded in real guideline and drug-label discrepancies - 2x2 factorial evaluation: Hint x Source design (NS, NS+Hint, SP, SP+Hint) - 56,000 evaluation records across 7 frontier LLMs - 400 expert-validated subset (TRIAGE-Bench-Verified) The 2x2 factorial design crosses two factors: GPT-5.6, Claude-Opus-4.8, GLM-5.2, Qwen-3.7, MiniMax-M3, DeepSeek-V4-Pro, Kimi-K2.6 1. Source dominates: Providing…