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.
Every trial of this one line, in one place: the 89-task main run, both re-run passes, and the scoring scripts. The trajectories are raw and unedited — each step's reasoning, each tool call, and the verifier's own stdout. This is a re-packaging, not a new measurement. The same files were published before, split across two releases, which made the line look incomplete in both: the first release carried the main run without the later re-runs, the second carried 11 re-runs without the main run. Nothing here is new data and no number changes. Both source releases remain up and unmodified. Three tiers, applied by one script (scripts/score4.py) identically to this and the three other lines it was…
Raw, unedited agent trajectories for the DeepSeek-V4.1-Flash run over all 89 Terminal-Bench 2.1 tasks, plus every remediation re-run for it and for the two older lines, plus the scoring and analysis scripts. Every number in the write-up can be traced to a file here. Exact McNemar, paired over the same 89 tasks: V4.1 is the only line above 70%, and it is 9 points above V4 on the same scaffold — but 89 tasks cannot carry that to significance. The direction is clear; "proven stronger" is not a claim this sample supports. Applied identically to every line, by one script (scripts/score4.py): 1. as-shipped — what the run actually scored. This is what you get if you deploy the same stack today. 2.…
LeRobot v3 dataset (albagon/telearmsdr3002looks): bimanual YAM teleoperation demos from the TeleArms game, re-rendered in MuJoCo with visual domain randomization for sim-to-real transfer of a pi0.5 policy. - Two independent DR looks per demo (dr00, dr01). The dr00 episodes are albagon/telearmsdr3001look. - 18 episodes whose wrist view was near-black for the whole episode are excluded (provenance/excludedarkwristtrain.json). - ckpt0's train/holdout split by demo UUID: provenance/ckpt0splitbyuuid.csv. The DR shares in the table below are for the whole render; the drreport summary covers all looks. Full config and per-factor ranges: provenance/drconfigresolved.json; render report…
This public archive preserves completed structured-latent-dynamics campaigns 03–07 from Pro6000 (WSL Ubuntu), together with the final local analyses, reports, audits, configurations, and source at the Git commit recorded in provenance.json. Each structured-latent-dynamics-campaignNN.tar.gz contains the corresponding directory formerly under /home/brand/, including results, checkpoints, raw records, generated labels, source snapshots, and agent working copies. Only the shared Python virtual environment and disposable Python/pytest caches are excluded. Installed package versions are in environment-freeze.txt. local-analyses-and-source.tar.gz contains tracked campaign 03–07 reports, audits…
This dataset was created using LeRobot.
ThunderWorld benchmark generated videos This release contains 33,880 generated MP4 videos for Cosmos3 Edge and Nano, evaluated on the full RBench and PAI-Bench scene inventories used by this study. Each model/method combination has two fixed seeds per scene: 650 RBench scenes and 1,044 PAI-Bench scenes. The files are generated outputs for downstream judging; judge scores are not included in this release. Download packages Each ZIP is independently usable and
This dataset was created using LeRobot.
A high-recall subset of datasocial/tiktok-5.6B-videos whose caption, hashtags or on-screen text mention a beverage or drink brand. - Partitioned by source month (data/month=YYYY-MM/) The source dataset is licensed under CC BY-NC 4.0. This subset inherits that license: non-commercial use only, with attribution to the source dataset.
This public pilot contains packed uint16 token IDs for pretraining the Tiny Base 300M, 2,048-context language model. It contains 50,000,000 training, 1,000,000 validation, and 1,000,000 test tokens. It does not redistribute raw documents. train.bin, validation.bin, and test.bin are little-endian uint16 streams. Use tokenizer.json with the Hugging Face tokenizers library. manifest.json records exact token counts, split policy, source revisions, mixture, and SHA-256 checksums; settings.json records preparation settings. The deterministic split is SHA-256 of normalized text modulo 1,000: 0–9 test, 10–19 validation, and 20–999 training. The mixture is 70% FineWeb-Edu, 25% English Wikipedia, and…
This is the official pretraining dataset for the TinyDTVi-500M causal language model. The corpus is aggressively filtered and deduplicated to maximize the representation of standard Vietnamese text for training resource-constrained models. The initial raw dataset (~135GB) was curated by mixing three high-quality open-source Vietnamese corpora. The approximate mixture ratio is: - 50% CulturaX-Vi (Cleaned web crawl) - 30% VNews (Vietnamese News Corpus) - 20% Vietnamese Wikipedia (Encyclopedia) After strict preprocessing and deduplication, the final dataset size is approximately 85GB (yielding roughly 18 billion unique tokens). To compress the vocabulary down to 50k tokens and focus strictly…
An Arena-style dataset of anonymized, multi-turn conversations focused on real-world tool use. It is intended for research, evaluation, and training of models that decide when and how to call tools. Data is organized into date-partitioned Parquet files. Each row contains one complete conversation trace with its tool calls and results. Flexible nested payloads are stored losslessly in metadatajson, inputjson, and outputjson; parse these JSON strings before training or analysis. processing.py provides streaming normalize and decode commands plus importable normalizetrace and decodetrace helpers for arbitrary JSON-compatible contexts. The repository is publicly discoverable, but files require…
This dataset was created using LeRobot.
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.