Removed objects (quality filter v1, 2026-10-05) 1,525 objects were removed after an automatic quality survey of their splats: untextured flat-grey objects (no colour information), and objects whose splats were mostly disconnected fragments, far-off floaters or oversized blurry Gaussians. Flat but strongly coloured objects were kept. removedqualityv1.csv lists every removed uid with the reason, the metric values and the paths it had. The
SAVRN Model Hub
AI Training Datasets
Each dataset with its card, its structure (every split, row count and column type), its files, its license, and the models that disclose training on it.
11,122Models
3,164Datasets
285Papers
6,792Publishers
11,852Sourced relationships
Updated 2026-10-18 · How the library is built
3,164 datasets, sorted by most downloaded.
Daimon-Infinity The world's largest omni-modal robotics dataset for physical AI, including high-resolution tactile sensing. To address the key bottleneck in fine manipulation data within the embodied intelligence field, Daimon has introduced the Daimon-Infinity dataset. The project will open-source over 10,000 hours of vision–tactile–language–action (VTLA) multimodal data to the entire industry. Its goal is to build a
Synthetic Nepali instruction-tuning data generated with NVIDIA NeMo Data Designer from authoritative Nepali documents (agriculture manuals, legal texts). Answers are grounded strictly in the source; unanswerable questions get an explicit refusal. Records use chat messages format plus metadata and per-record qualityscores (grounding / correctness / naturalness, 1-5, LLM-as-judge). One data/train-.jsonl per source document; shards are overwritten idempotently on re-runs. Records judged BELOW the quality gate; each has rejectreasons. Useful for judge calibration, hard-negative mining, or re-filtering with different thresholds. Do NOT use as-is for instruction tuning.
Synthetic topic models with a shared persona parent This dataset contains synthetic checkpoints and numerical evidence from the 10-hour persona/topic-model research sprint of 8–9 October 2026. Public archival was explicitly authorized by the owner after the private storage quota was exhausted. No natural-language training corpus or credentials are included. Read the research report and construction/EM proposal. The reproduction guide maps claims to code and numerical inputs. A
Every conversation behind every machine in the zot machinery - a software factory where an agent takes the same standing order every half hour, reads the catalogue of what already exists, looks at the last few machines on the shelf, and designs, writes, commissions and ships one brand-new machine: a working control panel for a machine that does not exist, with a live simulation behind it, faults you can trigger and recover from, and the operating manual that opens from the panel. Each row is one shift: the full agent trajectory from the order to the finished machine (or to where the shift was cut short), in the chat shape the rest of the ecosystem reads, plus what the machinery knows about…
Three full days of real crypto market microstructure (Binance spot), prepared the shape of the day (tick-by-tick relative price, normalized volumes, trade side, book imbalance) is fully preserved. The full feed — 27+ streams (raw L2 depth, 1-second trade tape, order-book metrics, derived features, regime labels) with SQL console, backtest runner and MCP access for AI agents — lives at data.cruscy.com. trades/: flow/: Data is recorded from the live exchange feed with sequence-gap auditing and
在 RoboTwin 2.0 里按官方轨迹做物理回放，导出每个 episode 前 132 帧 headcamera 的 ground-truth，供 OpenWAM 等表征的下游 probing 使用。 状态：已完成。 Agilex 2500 个 episode、Franka 2300 个 episode，共 4800 个 HDF5，全部通过结构检查，上传后都用 sha256 与 Hub 上的文件核对过。 episode 的范围就是 OpenWAM 仓库里 agilexrobotwin10smanifest.jsonl 和 frankarobotwin10smanifest.jsonl 列出的全部 episode，按 (task, episode) 对应。 episode.hdf5 里的数据都只来自 headcamera，没有 wrist、front 等其他相机，也没有视频和语言指令。手腕相机画面和其他补充数据在 episode extra.hdf5 里，见补充 GT。T 是该 episode 的帧数（不超过 132）。 - 分割：存的是调色板颜色，不是物体 ID，也没有颜色和物体名的对照表。颜色只在同一个 episode 里稳定，可以用来区分实例。上游 RoboTwin 的调色板只有 139 种颜色，遇到 label ID 更大的资产（例如 openlaptop）会报错，所以我们把调色板扩展到 256 种，前 139 种颜色保持原样。 - 几何检查：cam2worldgl 转成 OpenCV 约定后与 extrinsiccv 的差不超过 2e-7。把点云投影到同一帧的…
Versioned detector-training corpus. Current V1 repository contents: 87,568 rows, split 53,952 train / 12,787 validation / 20,829 test. The corpus stores source-specific provenance, hashes, grouping/de-duplication information, validation status and annotation metadata. It preserves the V1 training reference and provenance used by earlier strict FireViewer detector releases; each model receipt identifies the snapshot actually used. There is no single licence covering all source images. Keep repository-level license: other and preserve the source licence/attribution recorded for every sample. FireViewer publication must not overwrite source obligations. This corpus is a training/research…
prettyname: "IAAI Insurance Auto Auction Dataset" - open-data - data-science - web-scraping - scraper - crawler - data-collection - iaai - salvage-auction - insurance-auto-auction - total-loss - salvage-vehicles - automotive-data - vehicle-auction - parquet sizecategories: taskcategories: - other - configname: auction-listings datafiles: Sample of observed IAAI insurance auto auction lots with damage assessments, title status, bidding data, and branch locations. This dataset is a preview sample of the IAAI dataset published by Rebrowser. If you're doing academic research, you may be eligible for free access to a much larger slice — see Free Datasets for Research. This dataset contains 1…
Curated open intelligence dataset providing English briefs, clinical trial benchmarks, verified abstracts, and DOIs of frontier Chinese research in Cellular Therapeutics, Gene Editing, ADCs, and NMPA Clinical Approvals. If you utilize this open intelligence dataset in your academic research, industrial benchmarking, or LLM training pipelines, please cite our repository: Maintained automatically by Sino Open Intelligence Network.
Model Collections
Hand-picked starting points, each with the reason it exists.
Collection · 4 entries
Embedding models for retrieval
Sentence and document embedding models used to build retrieval systems. Dimension and sequence length matter more than size here, and both come from the publisher.
Collection · 4 entries
Models that fit on one accelerator
Models whose publisher-reported parameter count puts them within reach of a single accelerator at common precisions. Memory needed depends on precision and serving configuration, so treat the parameter count as the starting point, not the answer.
Collection · 6 entries
Open-weight text models worth knowing
Widely used open-weight language models, chosen because each one is a distinct family rather than a variant of the one above it. Selection, not a ranking.
Collection · 3 entries
Speech and audio models
Recognition and synthesis models, grouped so the two directions are easy to compare.
Related SAVRN Research
The hub sits beside SAVRN's market data and infrastructure research: what models cost to run, and what it takes to run them.
SAVRN Index
What open models cost to run
The same open-weight model priced by every host that serves it, per million tokens.
Research Hub
Data center trackers and maps
Moratoriums, permits, power, water and capital behind the facilities that run these models.
Method
How the Model Hub is built
Sources, evidence labels, refresh behaviour, and the limits of every comparison here.