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.
Results of vllm-sr/d3-edge at tag v3.0.2 = revision 5d17a013b2612f3bafe0d974356f35df835e4a36 on the complete Decision Index 0.3 public suite, run with the unmodified kit apolinario/decision-index 9eb2dbe (edition 0.3). all answered, 0 unsupported; scores.json says complete: true. - runs/d3-edge/: scores.json, index.json, benchmark-summary.json, results.jsonl.gz (the answers to every request), environment.json, status.json, RUNNOTES.md (engine, settings, hardware) and SHA256SUMS. - Re-scoring results.jsonl.gz with decisionindex score --edition 0.3 reproduces scores.json, index.json and benchmark-summary.json (apart from generatedutc).
Every file in this repository except this README is OpenPGP-encrypted (AES-256) to a single recipient key: fingerprint BE6C 5659 B9E4 D2A7 AB58 517A EE3F 3352 CB69 2630 Without the matching private key the contents cannot be read; nothing here is usable as published. Layout: large files are stored as.gpg.partNNN (8 GB pieces); directories of small files as.tar.gpg.partNNN. MANIFEST.json.gpg lists every item with its plaintext SHA-256 and part list. Restore (with the private key imported): cat.gpg.part | gpg -d >
A Japanese-language dataset for training and fine-tuning models on modern data engineering topics as of 2026. It contains instruction-tuning (SFT) Q&A pairs and a pretraining corpus covering distributed processing, streaming, data lakehouse, orchestration, transformation, local analytics, modeling, quality and governance, cloud and data platforms, MLOps, and AI/agentic data workflows. - AI / ML and data for AI: AI/LLM data pipelines (embeddings / RAG / vector DB), MLOps data (feature store / lineage / versioning), governance for AI agents (MCP / text2SQL) - Ops / FinOps / reliability, testing / CI, migration, design patterns All content reflects 2026-era trends (e.g. Iceberg as the open…
Instruction training data, organized by year (2013–2024).
Complete run of the full frozen Decision Index suite (150,759 requests, all answered ok) for - Median request latency 35.5 ms (1× RTX PRO 6000, HF Jobs) - Results stored with --compact (no suite text).
The DeepGlobe Road Extraction Dataset consists of 6,226 optical satellite images with a spatial dimension of $1024 \times 1024$ pixels and a Ground Sampling Distance (GSD) of 0.5 m/pixel. The imagery spans varied urban, suburban, and rural terrains across Thailand, Indonesia, and India. This repository hosts the standardized, reproducible benchmark partition: Partitioning is constructed deterministically with seed 42 from the sorted filename collection. If you use this dataset in your research, please cite the original challenge paper
Combined dataset for dental pathology detection on panoramic radiographs, in YOLO format. Built for training liodon-ai/dental-panoramic-detector. DENTEX provides: caries, periapical lesion, impacted tooth annotations on panoramic X-rays from 3 institutions. OralXrays-9 provides: decay (→ caries), apical periodontitis (→ periapicallesion), wisdom tooth (→ impactedtooth) annotations across a hospital-scale panoramic X-ray collection. Validation uses DENTEX val set only for consistent benchmarking. Then point your YOLO config data: to {path}/dental.yaml. If you use this dataset, please cite: This dataset combines and remaps annotations from the following sources — please also cite them
This dataset was created using LeRobot.
单条双臂遥操作 MCAP 录制切分出的 6 个 Episode 及抽帧结果,用于在 Colab 上验证 SmolVLM2-500M-Video-Instruct 能否替代 LLM 网关做 Episode 语义预标注。 含 /fsmstate 状态机、双臂末端位姿、夹爪关节、XR 手柄输入等信号。 - 原始录制:recording/rec202609141354110.mcap(约 36.4 秒,219935 条消息) - 切分口径:平台现有规则 episode.Method = fsmposev1,即优先按 /fsmstate 非零活动区间切分;本录制 6 个 Episode 全部命中 fsm 边界,置信度 0.90 - 帧抽取:每个 Episode 内按 frameCount 预算在 3 路视频流(左右彩相机 + XR 相机) 上均匀取样,JPEG 落盘 frames/index.json 每条形如: 帧按时间顺序编号。一个 Episode 取 8 帧,三路相机混合取样 (例如 episode-000 是 3 帧左相机 + 3 帧右相机 + 2 帧 XR 相机)。 平台 internal/pipelines/semanticepisodetag 活动当前把同样的证据交给 OpenAI 兼容视觉模型,产出 actionId / task / caption / outcome / confidence / needsHumanReview / evidenceFrameIds。这个数据集就是那条链路的输入快照, 用来先在 Colab…
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.