datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
GPT-5.5-Gemini-3.1-Pro-Grok-4-Claude-Fable-5-Mythos-5-Qwen-3.7-Max-and-more-Distillation-Dataset
📖 The Open Distillation Codex
🌌 The Ultimate Open-Source Distillation Dataset — No Skip, Full, with Attack & Defense 🌌
Where 73 open-source minds converge into one unified stream of intelligence
18M+ Distilled Signals · 7,090 Raw GitHub Repositories · 8 Curated Categories · ~76 GB+
"We did not write this dataset. We assembled it.
Every line is an echo — of a model thinking, a coder drafting, a tutor explaining, a repo breathing.
Seventy-three… See the full description on the dataset page: https://huggingface.co/datasets/Manusagents/GPT-5.5-Gemini-3.1-Pro-Grok-4-Claude-Fable-5-Mythos-5-Qwen-3.7-Max-and-more-Distillation-Dataset.project_gutenberg
Dataset Card for "Project Gutenberg"
Project Gutenberg is a library of over 70,000 free eBooks, hosted at https://www.gutenberg.org/.
All examples correspond to a single book, and contain a header and a footer of a few lines (delimited by a *** Start of *** and *** End of *** tags).
Usage
from datasets import load_dataset
ds = load_dataset("manu/project_gutenberg", split="fr", streaming=True)
print(next(iter(ds)))
License
Full license is available here:… See the full description on the dataset page: https://huggingface.co/datasets/manu/project_gutenberg.Infinity-Instruct
Infinity Instruct
Beijing Academy of Artificial Intelligence (BAAI)
[Paper][Code][🤗] (would be released soon)
The quality and scale of instruction data are crucial for model performance. Recently, open-source models have increasingly relied on fine-tuning datasets comprising millions of instances, necessitating both high quality and large scale. However, the open-source community has long been constrained by the high costs associated with building such extensive and… See the full description on the dataset page: https://huggingface.co/datasets/manifoldlabs/Infinity-Instruct.Omni-Frontier-Distillation-SFT-Cyber-Coding-Med-dataset-collection
🌌 Omni-Frontier Distillation SFT
The Definitive Evolution of Open-Source Distillation & Human-Crafted Expertise
Repository: Manusagents/Omni-Frontier-Distillation-SFT-Cyber-Coding-Med-dataset-collection
"The most comprehensive multi‑domain SFT corpus ever assembled — fusing 6.86 million cleaned distillation samples with 9.14 million human‑crafted expert examples across medical, cybersecurity, chemical, robotics, humanities, and more. 16 million… See the full description on the dataset page: https://huggingface.co/datasets/Manusagents/Omni-Frontier-Distillation-SFT-Cyber-Coding-Med-dataset-collection.corpus-1T-manifest
SPP Corpus 1T Manifest
The selection manifest for the ~1.0T-token pretraining corpus used in
Synthetic Persona Pretraining (SPP): Alignment from Token Zero.
The corpus is a seeded subsample of allenai/dolma3_mix-6T.
Rather than redistribute ~2.6 TB of text that is already public, this dataset
publishes the selection decisions keyed by upstream document id, so the corpus
can be reconstructed exactly by replaying against upstream.
📄 Reflections + text for the annotated half:… See the full description on the dataset page: https://huggingface.co/datasets/dlab-spp/corpus-1T-manifest.PulseLM
PulseLM: A Foundation Dataset and Benchmark for PPG-Text Learning
Usage
from datasets import load_dataset, get_dataset_config_names, concatenate_datasets # datasets==4.5.0
dataset_names = get_dataset_config_names("Manhph2211/PulseLM")
print(f"Available datasets: {dataset_names}")
train_splits = [
load_dataset("Manhph2211/PulseLM", name, split="train").select_columns(["signal", "text", "qa"])
for name in dataset_names
]
combined =… See the full description on the dataset page: https://huggingface.co/datasets/Manhph2211/PulseLM.Omni-Frontier-Distillation-SFT-Cyber-security-Coding-dataset-collection-v2
🧬 Omni-Frontier Collection
Cybersecurity · Coding · Math · Science · RSI Reasoning — one unified SFT package
A unified, deduplicated, fully-browsable distillation & SFT corpus — every row real, every row visible.
📖 Jump to
What's inside · 🔁 Aggregation audit · 🛡 Cybersecurity · 💻 Coding · 🏭 Distillation deep-dive · 🔁 RSI · 🧮 Math/Science/More · 🎓 Training guide · 🔎 Browsing · 🧹 Quality · 🗺 Roadmap · 📄 License… See the full description on the dataset page: https://huggingface.co/datasets/Manusagents/Omni-Frontier-Distillation-SFT-Cyber-security-Coding-dataset-collection-v2.manhattan
NextLat Manhattan Tokenized
Tokenized Manhattan random-walk dataset for world modeling evaluation of autoregressive models.
Each example stores a single pretokenized sequence in input_ids (list of int32), where the sequence format is:
<start_node> <end_node> <direction_1> ... <direction_n> end
Dataset structure
train/*.parquet: training split
heldout/*.parquet: heldout/eval split
manifest.json: split sizes and export metadata
tokenizer_meta.json: tokenizer vocabulary/id… See the full description on the dataset page: https://huggingface.co/datasets/JaydenTeoh/manhattan.python-copilot-training-from-many-repos-large
Python Copilot Large Coding Dataset
This dataset is a subset of the matlok python copilot datasets. Please refer to the Multimodal Python Copilot Training Overview for more details on how to use this dataset.
Details
Each row contains python code, either a class method or a global function, imported modules, base classes (if any), exceptions (ordered based off the code), returns (ordered based off the code), arguments (ordered based off the code), and more.
Rows: 2350782… See the full description on the dataset page: https://huggingface.co/datasets/matlok/python-copilot-training-from-many-repos-large.ManyIH-Bench
ManyIH-Bench
📄 Paper | 🏠 Homepage | 💻 Source Code
ManyIH-Bench is a benchmark for evaluating how well language models resolve instruction conflicts among instructions with arbitrarily many privilege levels. Each sample contains multiple instructions annotated with privilege markers; when instructions conflict, the model should follow the highest-privilege one.
The dataset has two subsets:
coding (427 samples): Code generation tasks (based on MBPP) with conflicting style… See the full description on the dataset page: https://huggingface.co/datasets/jhu-clsp/ManyIH-Bench.RAFE_Dataset
RAFE Dataset
This dataset repository stores the raw and intermediate data snapshots used by the public RAFE pipeline.
Contents
MDA_Raw/
Raw MD&A text files organized by ticker and filing form.
Extract/
Extracted factor artifacts, including:
*_TASK12_EXTRACTIONS.json
*_TASK13_AUDIT.json
*_FACTORS_DETAILED.json where available
Intended use
This repository is the raw-data and intermediate-artifact layer for the public RAFE workflow:
raw MD&A input
factor… See the full description on the dataset page: https://huggingface.co/datasets/NLPasset-mange/RAFE_Dataset.from-one-to-many-toxicity-mitigation
From One to Many: Expanding the Scope of Toxicity Mitigation in Language Models
[arxiv][code][data]
Data accompanying the paper "From One to Many: Expanding the Scope of Toxicity Mitigation in Language Models" accepted to ACL Findings 2024.
Abstract: To date, toxicity mitigation in language models has almost entirely been focused on single-language settings. As language models embrace multilingual capabilities, it’s crucial our safety measures keep pace. Recognizing this research… See the full description on the dataset page: https://huggingface.co/datasets/luizapzbn/from-one-to-many-toxicity-mitigation.Manu-FineWeb
Manu-FineWeb
Manu-FineWeb is a high-quality, large-scale corpus specifically curated for the manufacturing domain. It was extracted from the 15-trillion-token FineWeb dataset and refined to facilitate efficient domain-specific pretraining for models like ManufactuBERT.
Dataset Summary
Developed by: Robin Armingaud and Romaric Besançon (Université Paris-Saclay, CEA, List)
Statistics: 2B tokens/4,5 million documents
Construction & Curation
The dataset was… See the full description on the dataset page: https://huggingface.co/datasets/rarmingaud/Manu-FineWeb.verified-math-code-17k
Verified Math & Code, 17,000 rows
A math and code instruction dataset where every single row was mechanically checked before it was
allowed in. Not filtered by a heuristic, not scored by a model. Checked.
Two layers of verification, one per domain:
Every math answer was compared against an independent gold answer by exact, numeric and
symbolic (SymPy) comparison. If the worked solution did not arrive at the gold answer, the row
was dropped.… See the full description on the dataset page: https://huggingface.co/datasets/manifesta/verified-math-code-17k.brandvoice-marketing-briefs
BrandVoice Marketing Briefs
Generic AI writes like a robot. This trains it to write like a brand.
The loop is closed. The LoRA adapter
trained on this data scored +24.3% copy quality (7.0 to 8.7) with a 56% win rate against its
base, on Adaption's held-out judge. Dataset, weights, evaluation, and reproduction are all public.
A corpus of real marketing copy. 6,339 lines scraped from the live pages of 83 brands (the actual
Stripe, Liquid Death, Ramp, Duolingo copy)… See the full description on the dataset page: https://huggingface.co/datasets/manifesta/brandvoice-marketing-briefs.mantra
MANTRA
A deterministic, auditable programming language for ledger operations, cryptographic sealing, and verified integer computation.
Zero external dependencies. Pure Python. Auditable SHA-256. MUMPS gateway.
SnapKitty West / SNAPKITTYWEST — Evidence or Silence — 2026
What It Is
MANTRA is a purpose-built language for sovereign ledger operations. Every computation is deterministic and traceable. Every output can be cryptographically sealed. No external… See the full description on the dataset page: https://huggingface.co/datasets/Snapkitty/mantra.vuljector-extracted-vulnerabilities
Vuljector Extracted Vulnerabilities
This dataset is an export of the current Vuljector experiment artifacts.
Files:
produced_vulnerabilities.jsonl: one row per successfully produced injection attempt
verified_poc_successes.jsonl: one row per successful guided-PoC validation attempt
summary.json: aggregate counts
Current aggregate counts:
produced rows: 91
verified rows: 195
deduped unique valid cases: 267
Notes:
Counts are derived from all run artifacts under runs/injection… See the full description on the dataset page: https://huggingface.co/datasets/MananSuri27/vuljector-extracted-vulnerabilities.technical-manuals
Description
Topic: Technical Manuals
Domains: Engineering, Information Technology, Product Documentation
Number of Entries: 1,000
Dataset Type: Raw Dataset
Model Used: bedrock/us.meta.llama4-maverick-17b-instruct-v1:0
Language: English
Data-Analytics-Digital-Marketing-Project-Management-QA_DBhacking
Hacking Text Corpus
A research corpus of historical computer security writings, hacker zines, and hacktivist texts. Built for NLP, text generation, discourse analysis, and security research.
Contents
Phrack Magazine (phrack/)
72 issues (1985-2024), 1,026 articles
~55 MB of raw text, ~4.76 million words
Organized as phrack/issue{N}/{article}.txt
Topics: exploit development, reverse engineering, networking, phreaking, hacker culture, OS internals… See the full description on the dataset page: https://huggingface.co/datasets/Manusagents/hacking.ManipuriGPT-Corpus-v1.0
ManipuriGPT Corpus v1.0
ManipuriGPT Corpus v1.0 is a research-grade, multi-script, deduplicated, and quality-scored corpus specifically engineered for pretraining Manipuri (Meiteilon) language foundation models.
Quick Summary
Total Sequences: 147,956
Total Tokens (ManipuriGPT-Tokenizer-v1.0): 4,347,075
Total Characters: 16,019,401
Pipeline Version: 5.6
Release Version: v1.0.0
Build Timestamp: 2026-07-25T09:17:21.960438Z
Primary Writing Systems… See the full description on the dataset page: https://huggingface.co/datasets/nanskong/ManipuriGPT-Corpus-v1.0.mantinc-catalan-drift
Mantinc — Catalan Drift Benchmark
Descripció (ca)
Mantinc és un banc de proves que avalua si un model de llenguatge continua
responent en català quan el missatge, la conversa prèvia o el context recuperat
l'empenyen a fer-ho en una altra llengua, normalment el castellà o l'anglès.
Dataset Description
Mantinc is a benchmark that measures whether a language model keeps answering
in Catalan when the prompt, prior conversation, or retrieved context… See the full description on the dataset page: https://huggingface.co/datasets/softcatala/mantinc-catalan-drift.mantra
MANTRA
A deterministic, auditable programming language for ledger operations, cryptographic sealing, and verified integer computation.
Zero external dependencies. Pure Python. Auditable SHA-256. MUMPS gateway.
SnapKitty West / SNAPKITTYWEST — Evidence or Silence — 2026
What It Is
MANTRA is a purpose-built language for sovereign ledger operations. Every computation is deterministic and traceable. Every output can be cryptographically sealed. No external… See the full description on the dataset page: https://huggingface.co/datasets/SNAPKITTYWEST/mantra.titlegen-conversations
Combined Titlegen Conversations
This public release directly appends 10,284 accepted legacy title-generation
rows and 13,500 nine-language LLM-generated rows. The 23,784 examples are split
as train 20,584, validation 1,550, legacy test 200, legacy Vietnamese test 100,
and label-free synthetic holdout 1,350. Legacy rows contain only messages;
nine-language rows retain their richer IDs, language, coverage, cluster,
quality, and model-provenance fields. Train and validation… See the full description on the dataset page: https://huggingface.co/datasets/ManhHoDinh/titlegen-conversations.manifest-digital-identity-optimization
Manifest of Digital Identity Optimization (DIO) & Ontology of Digital Identity (ODI) — Hugging Face Distribution Layer
Version / Verze: 1.0.3 (Hugging Face Distribution Layer)
Author / Autor: Daniel Beránek
Date of public articulation / Datum veřejné artikulace: 2026-07-26
Primary public node / Primární veřejný uzel: https://danielberanek.cz/manifest-dio/
Canonical archival record / Kanonický archivní záznam: Zenodo, DOI: https://doi.org/10.5281/zenodo.21610934
License /… See the full description on the dataset page: https://huggingface.co/datasets/danielberanek/manifest-digital-identity-optimization.actuarial-global-glossary-multilingual
🤝 Connect with me on LinkedIn!
Join the mission to make actuarial knowledge accessible worldwide
Let's discuss how AI can transform professional education and break language barriers in finance!
🌍 Global Actuarial Glossary - Breaking Language Barriers in Finance
🚀 The World's Most Comprehensive Multilingual Actuarial Dataset
Imagine: A brilliant actuarial student in Tokyo, a risk analyst in São Paulo, and an insurance executive… See the full description on the dataset page: https://huggingface.co/datasets/manuelcaccone/actuarial-global-glossary-multilingual.manim-narrated-dpo-400
manim-narrated-dpo-400
Direct Preference Optimization (DPO) dataset pairing 361 verified, diverse narrated VoiceoverScene scripts (chosen) against structurally identical un-narrated silent Scene scripts (rejected), curated from authentic code-agent trajectories in nabin2004/AOS-Trajectories.
Dataset Summary
Size: 361 preference pairs (100% unique user visualization prompts).
Domains: Linear algebra (eigenvalues, SVD, transformations), calculus, machine learning… See the full description on the dataset page: https://huggingface.co/datasets/nabin2004/manim-narrated-dpo-400.verified-agronomy-17k
Verified Agronomy, 17,199 rows
The arithmetic a working agronomist actually does: fertiliser rates, growing degree days,
irrigation scheduling, seeding rates, grain moisture, sprayer calibration, forage budgeting.
Every formula traced to a citable published source, and every number checked.
And 1,796 rows (10.4%) that cannot be answered at all, where the correct response is to say so
and name the missing input.
Verify this card… See the full description on the dataset page: https://huggingface.co/datasets/manifesta/verified-agronomy-17k.tam-benchmarks
Tasks over Application Manuals (TAM)
TAM is a benchmark for evaluating long-horizon procedural reasoning: the ability of a language-model system to follow a large application manual, resolve cross-references, apply interdependent constraints, and produce an exact answer. Unlike short-horizon multi-hop tasks, TAM requires systems to maintain consistency across dozens of decisions drawn from manuals containing tens of thousands of rules. An early missed exception or incorrect… See the full description on the dataset page: https://huggingface.co/datasets/manulife/tam-benchmarks.manas-dataset-v2
Manas Dataset
Statistics
Total clean conversations: 1061
Train: 954
Eval: 107
Format
{
"conversations": [
{"from": "system", "value": "..."},
{"from": "human", "value": "..."},
{"from": "gpt", "value": "..."}
]
}