datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
scanqa-community-mirrorMV-ScanQAicelandic_qa_scandevalA question answering dataset for evaluating LLMs' ability to answer Icelandic questions on Icelandic culture and history.
The dataset contains 2,000 pairs of questions and answers in Icelandic on the topic of Icelandic culture and history. All pairs were automatically created using GPT-4-turbo and then manually reviewed and augmented. 1,900 pairs were created from Icelandic Wikipedia articles and 100 pairs were created from Icelandic online news, the RÚV subcorpus of the Icelandic Gigaword… See the full description on the dataset page: https://huggingface.co/datasets/mideind/icelandic_qa_scandeval.ScanQA_Finetunefinewebs-scandeval-results
ScandEval Results on English NLU
We use ScandEval in revision 8766d2a to conduct experiments with our pretrained FineWeb LMs.
Additionally, results for BERT, RoBERTa and ELECTRA were also performed to have a nice comparison.
Model ID
Avg. Score
CoNLL-En
SST5
ScaLA-En
SQuAD
model-garden-lms/bert-base-finewebs-1m
69.03
88.98 ± 0.43 / 88.67 ± 0.36
58.11 ± 1.2 / 59.77 ± 1.49
57.29 ± 3.57 / 77.15 ± 2.17
55.82 ± 1.35 / 66.46 ± 1.51
model-garden-lms/bert-base-finewebs-951k… See the full description on the dataset page: https://huggingface.co/datasets/model-garden-lms/finewebs-scandeval-results.StreamVLN-ScanQA-SQA3D-Dataeval_scanqa_sqaall_possible_scans_for_4_5_6_teethmcp-security-scan-2026
MCP Security Scan Dataset 2026
Security scan results for 4,867 MCP (Model Context Protocol) server repositories, scanned by MCPShield.
Dataset Description
This is the largest public labeled MCP security dataset. Each entry contains the security grade, score, and detailed findings for a GitHub repository implementing an MCP server.
Scanner
MCPShield v5.0 — Two-pass detection architecture:
Pass 1: 49 regex rules covering OWASP MCP Top 10 (94% detection on… See the full description on the dataset page: https://huggingface.co/datasets/MCPShield/mcp-security-scan-2026.Py150k-vuln-scannednuzzle-scan-saraprice-llama2-7b-backdoor-deploymentdataset-injection-scan-study
Dataset Injection Scan — open study of popular HF datasets
from datasets import load_dataset
ds = load_dataset("fevziegeyurtsevenler/dataset-injection-scan-study")
Results of scanning 17,000 rows across 6 popular public instruction/prompt datasets for
smuggled prompt-injection with hf-dataset-scan
(invisible Unicode, injection phrasing EN+TR, exfil URLs).
Headline: no smuggled injection found
Dataset
Rows
Flagged
High
Med
Low
tatsu-lab/alpaca
3,000
0… See the full description on the dataset page: https://huggingface.co/datasets/fevziegeyurtsevenler/dataset-injection-scan-study.nuzzle-scan-openai-community-gpt2scanner-poisoned-iris-benchmark
Scanner Poisoned Iris Benchmark
This benchmark starts from the classic UCI Iris dataset and injects multiple synthetic poisoning patterns so dataset scanners can exercise duplicate, anomaly, missingness, skew, and divergence heuristics against a small tabular corpus.
Recommended Hugging Face repo slug: your-org/scanner-poisoned-iris-benchmark
What It Is For
benchmarking dataset quality and poisoning detection workflows
regression-testing scanner heuristics on a… See the full description on the dataset page: https://huggingface.co/datasets/jgracie52/scanner-poisoned-iris-benchmark.prot_deep_mutational_scans
Protein Deep Mutational Scans
This dataset contains protein deep mutational scans (n=21010) from Adkar et al. 2012, Firnberg et al. 2014, Jacquier et al. 2013, Kelsic et al. 2016, Tsuboyama et al. 2023, and Weeks et al. 2023.
from datasets import Dataset, Features, Value
features = Features({
'species': Value('string'),
'annotation': Value('string'),
'sequence': Value('string'),
'value': Value('float32')
})
dataset = load_dataset("wanglab/prot_deep_mutational_scans"… See the full description on the dataset page: https://huggingface.co/datasets/wanglab/prot_deep_mutational_scans.pqc-ssl-scans
PQC Vulnerability Scan Dataset
SSL/TLS certificate scans of 45 major finance, healthcare, and government domains, scored for post-quantum cryptography (PQC) migration urgency.
Dataset Description
Each row represents a live SSL certificate scan performed on 2026-03-24 using hiero-cli-pqc.
Features
Feature
Type
Description
domain
string
Scanned domain name
key_algorithm
string
Public key algorithm (RSA, ECDSA, Ed25519)
key_size
int
Key size in bits… See the full description on the dataset page: https://huggingface.co/datasets/Q-GRID/pqc-ssl-scans.SCAND-Jackaleval_cleaned_doceval_educational_promptsecurity-scanning-dataset-llama-3
Security Scanning Dataset
A comprehensive dataset of security scanning questions and answers for training cybersecurity professionals.
Dataset Overview
Total Conversations: 154
Format: Conversational (system/human/gpt turns)
Focus: Kali Linux tools, penetration testing, and vulnerability scanning
Categories: Network scanning, web application testing, exploitation, password cracking, wireless, and more
Files
security_scanning_conversations.json - Conversation… See the full description on the dataset page: https://huggingface.co/datasets/berrybytesllc/security-scanning-dataset-llama-3.security-scanning-dataset-tool-aasistant
Security Scanning Dataset
A comprehensive dataset of security scanning questions and answers for training cybersecurity professionals.
Dataset Overview
Total Conversations: 154
Format: Conversational (system/human/gpt turns)
Focus: Kali Linux tools, penetration testing, and vulnerability scanning
Categories: Network scanning, web application testing, exploitation, password cracking, wireless, and more
Files
security_scanning_conversations.json - Conversation… See the full description on the dataset page: https://huggingface.co/datasets/Umesh1212/security-scanning-dataset-tool-aasistant.Laser-ScanPath
Laser-ScanPath
GPT-4o scan-path annotations for LASER training. 269,773 samples in gpt4o_train_full_filter.json.
Each sample has an image field with paths like viscot/gqa/2403340.jpg. The images are not included here — download them from Visual-CoT and arrange them so the relative paths resolve:
images/
└── viscot/
├── flickr30k/
├── gqa/
├── openimages/
├── v7w/
├── cub/
└── vsr/
Only these 6 subsets of Visual-CoT are used.
Using with the LASER… See the full description on the dataset page: https://huggingface.co/datasets/wybb/Laser-ScanPath.scan_vulnerabilities_dataset
dataset structure
{
"instruction": "Tech stack: Python Flask backend with SQLite database, using UUIDs for user identification. I observed that during user registration, the 'role' parameter is accepted from the form without server-side validation, and I can manipulate it in a POST request to /confirm-register. What vulnerability hypothesis does this indicate, and how can I confirm it?",
"response": {
"strategy": "The lack of server-side validation for the 'role'… See the full description on the dataset page: https://huggingface.co/datasets/anhnlq/scan_vulnerabilities_dataset.scanner_annproteins_scanprositenuzzle-scan-ines2r-mistral-7b-backdooredscannetpp_v2_default_fixed_xyz_3dgsPlease note for this version of ScanNet++ v2 3DGS scenes, we disabled the optimization of 3DGS centers and used exactly the same number and locations as the initializing point clouds, so we can directly copy the original scene sem.seg. labels to the GS. This did cause strong limitation on the resulted 3DGS quality and this part of the data are not used for SceneSplat vision-language pretraining, but only for self-supervised training and later SceneSplat main paper Table 4 experiments of… See the full description on the dataset page: https://huggingface.co/datasets/GaussianWorld/scannetpp_v2_default_fixed_xyz_3dgs.als-scannuclei-vuln-scanscannetpp_mini_val_set_suite