datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
dolmino-mix-1124
DOLMino dataset mix for OLMo2 stage 2 annealing training.
Mixture of high-quality data used for the second stage of OLMo2 training.
Source Sizes
Name
Category
Tokens
Bytes (uncompressed)
Documents
License
DCLM
HQ Web Pages
752B
4.56TB
606M
CC-BY-4.0
Flan
HQ Web Pages
17.0B
98.2GB
57.3M
ODC-BY
Pes2o
STEM Papers
58.6B
413GB
38.8M
ODC-BY
Wiki
Encyclopedic
3.7B
16.2GB
6.17M
ODC-BY
StackExchange
CodeText
1.26B
7.72GB
2.48M
CC-BY-SA-{2.5, 3.0, 4.0}… See the full description on the dataset page: https://huggingface.co/datasets/allenai/dolmino-mix-1124.dolma3_mix-150B-1025
Dolma 3 Sample: 150B Mix
Dataset Sources
Sample of data for 1Bx5C and 7Bx1B. For the full Dolma 3 pool, see: https://huggingface.co/datasets/allenai/dolma3
Source
Type
Tokens
Documents
Common Crawl
Web pages
121B (76.9%)
84.5M
olmOCR Science PDFs
Academic documents
19.9B (12.6%)
2.25M
Stack-Edu (Rebalanced)
GitHub code
11.1B (7.06%)
14.3M
arXiv
Papers with LaTeX
1.29B (0.82%)
247K
FineMath 3+
Math web pages
4.10B (2.60%)
2.57M
Wikipedia & Wikibooks… See the full description on the dataset page: https://huggingface.co/datasets/allenai/dolma3_mix-150B-1025.tulu-2.5-preference-data
Tulu 2.5 Preference Data
This dataset contains the preference dataset splits used to train the models described in Unpacking DPO and PPO: Disentangling Best Practices for Learning from Preference Feedback.
We cleaned and formatted all datasets to be in the same format.
This means some splits may differ from their original format.
To see the code used for creating most splits, see here.
If you only wish to download one dataset, each dataset exists in one file under the data/… See the full description on the dataset page: https://huggingface.co/datasets/allenai/tulu-2.5-preference-data.ALLaVA-4V
📚 ALLaVA-4V Data
Generation Pipeline
LAION
We leverage the superb GPT-4V to generate captions and complex reasoning QA pairs. Prompt is here.
Vison-FLAN
We leverage the superb GPT-4V to generate captions and detailed answer for the original instructions. Prompt is here.
Wizard
We regenerate the answer of Wizard_evol_instruct with GPT-4-Turbo.
Dataset Cards
All datasets can be found here.
The structure of naming is shown below:
ALLaVA-4V
├──… See the full description on the dataset page: https://huggingface.co/datasets/FreedomIntelligence/ALLaVA-4V.ALL-Bench-Leaderboard
🏆 ALL Bench Leaderboard 2026
The only AI benchmark dataset covering LLM · VLM · Agent · Image · Video · Music in a single unified file.
Dataset Summary
ALL Bench Leaderboard aggregates and cross-verifies benchmark scores for 90+ AI models across 6 modalities. Every numerical score is tagged with a confidence level (cross-verified, single-source, or self-reported) and its original source. The dataset is designed for researchers, developers, and… See the full description on the dataset page: https://huggingface.co/datasets/FINAL-Bench/ALL-Bench-Leaderboard.All-CVE-Records-Training-Dataset
CVE Chat‑Style Multi‑Turn Cybersecurity Dataset (1999 – 2025)
1. Project Overview
This repository hosts the largest publicly available chat‑style, multi‑turn cybersecurity dataset to date, containing ≈ 300 000 Common Vulnerabilities and Exposures (CVE) records published between 1999 and 2025. Each record has been meticulously parsed, enriched, and converted into a conversational format that is ideal for training and evaluating AI and AI‑Agent systems focused on… See the full description on the dataset page: https://huggingface.co/datasets/AlicanKiraz0/All-CVE-Records-Training-Dataset.tutormoments-preview
TutorMoments-Preview
462 real K–12 math tutoring sessions (student and tutor) with human annotations, plus a benchmark of
7,280 AI-tutor attempts scored the same way. A preview release from TutorMoments, a project on how well
tutors — human and AI — scaffold, push for rigor, and build rapport. From one K–12 tutoring program
(anonymized as tutoring_provider_a).
Paper: When Help is Unhelpful: Evaluating AI Tutors for Productive Struggle
Code:… See the full description on the dataset page: https://huggingface.co/datasets/allenai/tutormoments-preview.ALL-Bench-Leaderboard
🏆 ALL Bench Leaderboard 2026
The only AI benchmark dataset covering LLM · VLM · Agent · Image · Video · Music in a single unified file.
Dataset Summary
ALL Bench Leaderboard aggregates and cross-verifies benchmark scores for 90+ AI models across 6 modalities. Every numerical score is tagged with a confidence level (cross-verified, single-source, or self-reported) and its original source. The dataset is designed for researchers, developers, and… See the full description on the dataset page: https://huggingface.co/datasets/youssef3146/ALL-Bench-Leaderboard.CNY-data
CNY walk-ready graph reasoning data
Walk-ready data for CNY (Call Neighbours Yourself), the graph-walk reinforcement
learning framework described in Call Neighbours Yourself: Graph Walks with
Destination-Conditioned On-Policy Self-Distillation (EMNLP 2026), arXiv:2608.29588.
Models: Allen-UQ/CNY-7B,
Allen-UQ/CNY-14B.
This repository does not redistribute any graph dataset in its original form. It
provides the derived, prompt-rendered form in which each graph instance is… See the full description on the dataset page: https://huggingface.co/datasets/Allen-UQ/CNY-data.commongen_lite
CommonGen-Lite
Evaluating LLMs with CommonGen using CommonGen-lite dataset (400 examples + 900 human references).
We use GPT-4 to evaluate the constrained text generation ability of LLMs. Please see more in our paper.
Github: https://github.com/allenai/CommonGen-Eval
Leaderboard
model
len
cover
pos
win_tie
overall
human
12.84
99.00
98.11
100.00
97.13
gpt-4-0613
14.13
97.44
91.78
50.44
45.11
gpt-4-1106-preview14.90
96.33
90.11
50.78
44.08… See the full description on the dataset page: https://huggingface.co/datasets/allenai/commongen_lite.allenai-WildChat
AllenAI WildChat Combined Dataset
This unofficial repository provides the AllenAI WildChat Combined Dataset, which merges the WildChat-4.8M and WildChat-1M collections of human–ChatGPT conversations.
WildChat-1M contains 1 million chats, of which 25.53% are from GPT‑4 and the remainder from GPT‑3.5. These conversations cover a wide range of complex interactions, including code-switching, ambiguity, and political topics.
WildChat-4.8M originally comprised 4.8 million conversations.… See the full description on the dataset page: https://huggingface.co/datasets/agentlans/allenai-WildChat.ALLaVA-4V
📚 ALLaVA-4V Data
Generation Pipeline
LAION
We leverage the superb GPT-4V to generate captions and complex reasoning QA pairs. Prompt is here.
Vison-FLAN
We leverage the superb GPT-4V to generate captions and detailed answer for the original instructions. Prompt is here.
Wizard
We regenerate the answer of Wizard_evol_instruct with GPT-4-Turbo.
Dataset Cards
All datasets can be found here.
The structure of naming is shown below:
ALLaVA-4V… See the full description on the dataset page: https://huggingface.co/datasets/lodestones/ALLaVA-4V.allenai-WildChat-4.8MAll-CVE-Chat-MultiTurn-1999-2025-Dataset
CVE Chat‑Style Multi‑Turn Cybersecurity Dataset (1999 – 2025)
1. Project Overview
This repository hosts the largest publicly available chat‑style, multi‑turn cybersecurity dataset to date, containing ≈ 300 000 Common Vulnerabilities and Exposures (CVE) records published between 1999 and 2025. Each record has been meticulously parsed, enriched, and converted into a conversational format that is ideal for training and evaluating AI and AI‑Agent systems focused on… See the full description on the dataset page: https://huggingface.co/datasets/Trendyol/All-CVE-Chat-MultiTurn-1999-2025-Dataset.allenai-dolma3_mix-6T-sampletulu-v2-sft-mixture-olmo-2048
Dataset Card for Tulu V2 Mix (2048 OLMo version)
Note the ODC-BY license, indicating that different licenses apply to subsets of the data. This means that some portions of the dataset are non-commercial. We present the mixture as a research artifact.
This is a modified version of the Tulu V2 Mix used to train OLMo-Instruct.
The two primary differences are: long conversations are resplit into 2048-token chunks, and the hardcoded subset has been replaced with similar examples about… See the full description on the dataset page: https://huggingface.co/datasets/allenai/tulu-v2-sft-mixture-olmo-2048.EnvFactory-SFT-ALL
EnvFactory: Scaling Tool-Use Agents via Executable Environments Synthesis and Robust RL
## Overview
EnvFactory-SFT-ALL is the complete supervised fine-tuning (SFT) dataset containing 26,500 tool-use trajectories synthesized using the EnvFactory framework. This dataset includes all generated trajectories before filtering.
The dataset contains multi-turn tool-use trajectories with implicit human reasoning, generated through… See the full description on the dataset page: https://huggingface.co/datasets/LARK-Lab/EnvFactory-SFT-ALL.allenai-WildChat-4.8M-prompts
allenai/WildChat-4.8M English Prompts
Dataset Summary
This dataset contains real user-submitted prompts to ChatGPT, extracted from the English portion of the allenai/WildChat-4.8M collection.
It serves as a large-scale resource for analyzing user intent, conversational diversity, and prompt engineering patterns.
Files
en_prompts: All English-language first messages from user conversations.
Each record represents the first user prompt.
Exact duplicates are… See the full description on the dataset page: https://huggingface.co/datasets/agentlans/allenai-WildChat-4.8M-prompts.forbidden-backrooms-gemma-4-31B-it
Forbidden Backrooms: Gemma-4 31B Self-Chat
Self-chat transcripts and per-message embeddings for two role-inverted instances of Gemma-4-31B-it, comparing the official instruct checkpoint against an abliterated fine-tune of the same checkpoint. Both variants use identical int4 quantization served via Ollama, so quantization noise is not a confound between them.
The methodology follows Anthropic's Claude Opus 4 system card section on the "spiritual bliss attractor state." Leave two… See the full description on the dataset page: https://huggingface.co/datasets/alliedtoasters/forbidden-backrooms-gemma-4-31B-it.ALLaVA-4V-Chinese
ALLaVA-4V for Chinese
This is the Chinese version of the ALLaVA-4V data. We have translated the ALLaVA-4V data into Chinese through ChatGPT and instructed ChatGPT not to translate content related to OCR.
The original dataset can be found here, and the image data can be downloaded from ALLaVA-4V.
Citation
If you find our data useful, please consider citing our work! We are FreedomIntelligence from Shenzhen Research Institute of Big Data and The Chinese University of… See the full description on the dataset page: https://huggingface.co/datasets/FreedomIntelligence/ALLaVA-4V-Chinese.all-cve-chat-multiturn-1999-2025
CVE Chat‑Style Multi‑Turn Cybersecurity Dataset (1999 – 2025)
1. Project Overview
This repository hosts the largest publicly available chat‑style, multi‑turn cybersecurity dataset to date, containing ≈ 300 000 Common Vulnerabilities and Exposures (CVE) records published between 1999 and 2025. Each record has been meticulously parsed, enriched, and converted into a conversational format that is ideal for training and evaluating AI and AI‑Agent systems focused on… See the full description on the dataset page: https://huggingface.co/datasets/ansulev/all-cve-chat-multiturn-1999-2025.tulu-v2-sft-mixture-olmo-4096
Dataset Card for Tulu V2 Mix (4096 OLMo version)
Note the ODC-BY license, indicating that different licenses apply to subsets of the data. This means that some portions of the dataset are non-commercial. We present the mixture as a research artifact.
This is a modified version of the Tulu V2 Mix used to train newer (after April 2024) OLMo-SFT/Instruct variants (e.g. this model, or this one).
The only difference is that the hardcoded subset (dataset='hard_coded') has been replaced… See the full description on the dataset page: https://huggingface.co/datasets/allenai/tulu-v2-sft-mixture-olmo-4096.allenai_WildChat-1M-gemini-2.0-flash-thinking-exp-1219-CustomShareGPT
allenai_WildChat-1M-gemini-2.0-flash-thinking-exp-1219-CustomShareGPT
PJMixers-Dev/allenai_WildChat-1M-prompts with responses generated with gemini-2.0-flash-thinking-exp-1219.
Generation Details
If BlockedPromptException, StopCandidateException, or InvalidArgument was returned, the sample was skipped.
If ["candidates"][0]["safety_ratings"] == "SAFETY" the sample was skipped.
If ["candidates"][0]["finish_reason"] != 1 the sample was skipped.
model =… See the full description on the dataset page: https://huggingface.co/datasets/PJMixers-Dev/allenai_WildChat-1M-gemini-2.0-flash-thinking-exp-1219-CustomShareGPT.allaM-offsec-arabic-chat-v2
Arabic Offensive Security Chat Dataset v2
High-quality category-aware bilingual Arabic/English dataset for offensive security assistants.
What's New in v2
✅ Category-aware responses: Different response structures for web vulns, DeFi, reconnaissance tools, social engineering, etc.
✅ No generic templates: Each category has specialized analysis framework
✅ No verbatim copying: Responses analyze and transform the input, not repeat it
✅ Semantic accuracy: Tools (nmap… See the full description on the dataset page: https://huggingface.co/datasets/haiderkamal23/allaM-offsec-arabic-chat-v2.tulu-v2-sft-long-mixtureThis is a recreation of the tulu-v2-sft-mixture, without splitting ShareGPT dataset into chunks of max 4096 tokens. This might be interesting to people who are doing long-context finetuning.
Please refer to the original tulu-v2-sft-mixture for the details of this dataset mixture.
License
We are releasing this dataset under the terms of ODC-BY. By using this, you are also bound by the Common Crawl terms of use in respect of the content contained in the dataset.
All-CVE-Records-Training-Dataset-archive
CVE Chat‑Style Multi‑Turn Cybersecurity Dataset (1999 – 2025)
1. Project Overview
This repository hosts the largest publicly available chat‑style, multi‑turn cybersecurity dataset to date, containing ≈ 300 000 Common Vulnerabilities and Exposures (CVE) records published between 1999 and 2025. Each record has been meticulously parsed, enriched, and converted into a conversational format that is ideal for training and evaluating AI and AI‑Agent systems focused on… See the full description on the dataset page: https://huggingface.co/datasets/ChipHolmes/All-CVE-Records-Training-Dataset-archive.ALLaVA-4V-Arabic
ALLaVA-4V for Arabic
This is the Arabic version of the ALLaVA-4V data. We have translated the ALLaVA-4V data into Arabic through ChatGPT and instructed ChatGPT not to translate content related to OCR.
The original dataset can be found here, and the image data can be downloaded from ALLaVA-4V.
Citation
If you find our data useful, please consider citing our work! We are FreedomIntelligence from Shenzhen Research Institute of Big Data and The Chinese University of Hong… See the full description on the dataset page: https://huggingface.co/datasets/FreedomIntelligence/ALLaVA-4V-Arabic.CompCert-eval
CompCert Proof-Engineering Eval
Proof-synthesis challenges mined from the git history of
AbsInt/CompCert, the formally verified C
compiler. Each challenge is a real proof-engineering edit that a human made in a
single commit: we take the repository state before the commit (the
challenge) and treat the state after the commit (the solution) as
ground truth. The model's job is to reconstruct the proof/spec work the human
did.
⚠️ License notice. CompCert is distributed under the… See the full description on the dataset page: https://huggingface.co/datasets/for-all-dev/CompCert-eval.allenai-soda
SODA chat dataset
This is the SODA dataset in ShareGPT-like format.
According to the dataset's creators: "🥤SODA is the first publicly available, million-scale, high-quality dialogue dataset covering a wide range of social interactions."
The following changes were made to the data:
kept only dialogues with two people alternating turns
the SODA narrative was adapted into a system prompt for an AI to roleplay as the second person
extra turns were removed so that each conversation… See the full description on the dataset page: https://huggingface.co/datasets/agentlans/allenai-soda.All-CVE-Chat-MultiTurn-1999-2025-Dataset
CVE Chat‑Style Multi‑Turn Cybersecurity Dataset (1999 – 2025)
1. Project Overview
This repository hosts the largest publicly available chat‑style, multi‑turn cybersecurity dataset to date, containing ≈ 300 000 Common Vulnerabilities and Exposures (CVE) records published between 1999 and 2025. Each record has been meticulously parsed, enriched, and converted into a conversational format that is ideal for training and evaluating AI and AI‑Agent systems focused on… See the full description on the dataset page: https://huggingface.co/datasets/ukcli/All-CVE-Chat-MultiTurn-1999-2025-Dataset.