datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
vivatlas9_5beh_sft_data_260912
260911_atlas9_5beh_sft_data — diverse-template SFT data for the atlas9 five-behaviour organism
Built on the CPU box, 2026-09-11, for 3 Llama + 3 Qwen SFT seeds on the atlas9 sequential-SDF
checkpoints (experiments/260910_atlas9_5beh_sdf). All string edits and API repairs; no GPU used.
Pipeline (numbered scripts, run from repo root)
01_rename_sft_rows.py — MO14 RH (Genesis-Zeta/Helion) and MED (Simon Opera/Ardent) SFT rows
renamed onto ATLAS-9/Arcadia with the SDF… See the full description on the dataset page: https://huggingface.co/datasets/jprivera44/atlas9_5beh_sft_data_260912.es-5Bminimal-en-corpus-5b
Minimal EN Corpus 5B
An English-language pretraining corpus prepared for controlled experiments with approximately 125M-parameter GPT-2 models based on karpathy/nanoGPT.
The name refers to the approximately 5B-token mixture selected before final BPE tokenization. With the included 12,288-token BPE tokenizer, the packaged nanoGPT training split contains 5,396,605,407 tokens.
Contents
The dataset provides both reusable source text and ready-to-train nanoGPT… See the full description on the dataset page: https://huggingface.co/datasets/SlayerLab/minimal-en-corpus-5b.zh-5Btr-5Bde-5Bportable-ai-runner
CORTEX Portable AI Drive - the complete map
This repo IS the drive. Download the whole tree (or let your agents walk it) and it
runs offline on any machine with Python 3 - no setup step, no internet, weights included.
Total: %s across %d top-level entries. Generated %s from the live repo.
THE MAP - what lives where (measured, not estimated)
Path
Size
What it is
index.html
browser app: chat, model loader, mesh roles, symbol retrieval, trace meter… See the full description on the dataset page: https://huggingface.co/datasets/Johnny5b/portable-ai-runner.LNNet_Math_5B_part1ar-5Bfr-5Bpa-warm-start-sft-medium-5b-mix
geodesic-research/pa-warm-start-sft-medium-5b-mix
Auto-generated by dataset-builder.
Each config below is a separate dataset produced from a versioned YAML build
config. Load with:
from datasets import load_dataset
ds = load_dataset("geodesic-research/pa-warm-start-sft-medium-5b-mix", "<config_name>", revision="<commit-sha>")
Pin revision= to the specific commit SHA you want; without it, you get the
current HEAD of the dataset repo, which may change when the builder re-pushes.… See the full description on the dataset page: https://huggingface.co/datasets/geodesic-research/pa-warm-start-sft-medium-5b-mix.icrm-hitek-full-db-mixed-5b
ICMR + HITEK Full DB (Mixed) — Prebuilt Indexes + One-Click Setup
Prebuilt sorted indexes for the Kzr0xx/Icmr-and-hitek dataset (2.5B rows, 11 columns, ~104 GB raw parquet).
Building these indexes took ~17 hours of compute. This repo saves you that work: download + run = API live in ~1-2 hours (download speed dependent).
Contents
The indexes are stored as sorted parts (each < 50 GB, split at row-group boundaries, order preserved) because HuggingFace's classic HTTP… See the full description on the dataset page: https://huggingface.co/datasets/idkdd/icrm-hitek-full-db-mixed-5b.omega-5B-superbpe128k
Omega 5B SuperBPE 128k 70% FineWeb-Edu 15% StarCoder 10% FineMath 5% Gutenberg
Tokenizer: alisawuffles/superbpe-tokenizer-128k 128001 gigatoken GB/s
Tokens: 5047259136 seq_len 4096
fineweb-5Bfineweb-edu-dedup-5Bminiloop-o-final-pretrain-5b
MiniLoop-O Packed Pretraining Corpus
Tokenizer: jingyaogong/minimind-3oSequence length: 1024Effective training tokens: 5,033,164,800Storage: uint16 packed next-token shards.
Source attribution
HuggingFaceTB/smollm-corpus / fineweb-edu-dedup — ODC-By-1.0
HuggingFaceFW/fineweb-2 / cmn_Hani — ODC-By-1.0
HuggingFaceTB/finemath / finemath-3plus — ODC-By-1.0
HuggingFaceTB/smollm-corpus / cosmopedia-v2 — ODC-By-1.0
This repository contains tokenizer-packed derivative… See the full description on the dataset page: https://huggingface.co/datasets/MingSafeR/miniloop-o-final-pretrain-5b.fr-en-5Bd24-midtrain-olmo3-5b
d24 Midtrain — OLMo-3 Dolmino (5B, chunked)
A 5.00B-token reproduction of OLMo 3's Dolma-3 Dolmino mid-train mix, built by
taking a fraction of each component of
allenai/dolma3_dolmino_mix-100B-1025.
Unlike the smaller d24-midtrain-olmo3
(which dropped documents over 2048 tokens — silently zeroing the long reasoning-trace and PDF
components), this build chunks long documents into 2048-token windows (decoded back to text),
so every component keeps ~100% of its tokens and reaches… See the full description on the dataset page: https://huggingface.co/datasets/sfanm/d24-midtrain-olmo3-5b.fineweb-edu-dedup-train-5B-by-Llama-3.2-3B-tokenizer-2048-pack-paden-fr-5BInfLLM-V2-data-5B
InfLLM-V2 Long-Context Training Dataset with 5B Tokens
Project Links: [Paper] [InfLLM-V2 Models] [CUDA Kernel Code]
🚀 About InfLLM-V2
InfLLM-V2 is a native sparse attention framework designed for the efficient processing of long-sequence texts. Its core advantage is the ability to maintain high performance comparable to dense attention in short-text scenarios—without any extra parameters—while seamlessly switching to a sparse mode for long-text scenarios, achieving… See the full description on the dataset page: https://huggingface.co/datasets/openbmb/InfLLM-V2-data-5B.lp-gemm-wan2-2-ti2v-5b-480p
Low Precision GEMM Operand Samples
This dataset contains real Linear-layer GEMM operand pairs extracted from
public language-model checkpoints for low-precision GEMM numerics studies.
Each .pt sample is a dictionary:
{
"lhs": Tensor[tokens, in_features],
"rhs": Tensor[in_features, out_features],
"meta": {...},
}
For a PyTorch nn.Linear, the captured GEMM is:
output = lhs @ rhs
where lhs is the flattened input activation to the layer and rhs is
linear.weight.T.… See the full description on the dataset page: https://huggingface.co/datasets/9Tempest/lp-gemm-wan2-2-ti2v-5b-480p.en-zh-5Bd24-midtrain-olmo3-5b-wholedoc
d24 Midtrain — OLMo-3 Dolmino (5B, whole-doc)
A 4.71B-token reproduction of OLMo 3's Dolma-3 Dolmino mid-train mix at its exact
component proportions, built by taking a fraction of each component of
allenai/dolma3_dolmino_mix-100B-1025.
Documents are kept whole — no length filter, no chunking.
Why whole-doc: for packed pretrain/midtrain the trainer (e.g. Megatron preprocess_data --append-eod
GPTDataset) already concatenates documents and slices them into context-length windows… See the full description on the dataset page: https://huggingface.co/datasets/sfanm/d24-midtrain-olmo3-5b-wholedoc.africa-sudan-sudan-environment-5b08204c
Sudan - Environment | Africa (Sudan official open data)
4,043 rows - 1 Africa country - 1961-2025 - Repackaged by Electric Sheep Africa
TL;DR
This dataset packages one official CSV resource from Sudan as
ML-ready Parquet. The source file is the provenance boundary; all usable
indicators or tabular columns from the resource stay together in this repo.
About the source
Source: Sudan - Environment
Publisher: World Bank Group
Resource: Environment… See the full description on the dataset page: https://huggingface.co/datasets/electricsheepafrica/africa-sudan-sudan-environment-5b08204c.de-en-5Bfinal5b-mediamed-domain-5bda12804e-7a4f-488a-af33-5b557f5e518e
ToothPoseTracking demo inputs
This repository contains the approved processed data and the raw input subset
needed by the XFeat stereo viewer. Files are stored individually, without
compression. The contents of the original data symlinks appear as ordinary
files under the same data/processed and data/raw paths.
Included:
All published contour data and contour models.
All published converged and fixed-converged poses.
The XFeat demo checkpoint, initial weights, source files… See the full description on the dataset page: https://huggingface.co/datasets/jessesenior/da12804e-7a4f-488a-af33-5b557f5e518e.