datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
lingshu_training_data_medical_domain
Website
🤖 7B Model
🤖 8B Model based on InternVL3
🤖 32B Model
MedEvalKit
Technical Report
Lingshu MCP
Lingshu Medical MLLM Training Data (Medical Domain)
This dataset contains the medical-domain training data used in the multi-stage training of the Lingshu Medical Multimodal Large Language Model (MLLM). General-domain data has been removed; only medical data is included.
The training… See the full description on the dataset page: https://huggingface.co/datasets/lingshu-medical-mllm/lingshu_training_data_medical_domain.Swallow-Nemotron-Post-Training-Dataset-v1
Swallow-Nemotron-Post-Training-Dataset-v1
The Swallow LLM Project constructed the Swallow-Nemotron-Post-Training-Dataset-v1 based on the math, code, and stem subsets of the NVIDIA Nemotron-Post-Training-Dataset-v1, as illustrated in the figure below.
Dataset Construction
The original Thinking Trajectories and Assistant Outputs in the Nemotron-Post-Training-Dataset-v1 were synthesized using DeepSeek-R1-0528.
However, we identified an issue with the Thinking… See the full description on the dataset page: https://huggingface.co/datasets/tokyotech-llm/Swallow-Nemotron-Post-Training-Dataset-v1.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.AgentDoG1.0-Training-Data
AgentDoG1.0 Training Data
[💻 GitHub] | [📊 ATBench Dataset] | [📄 ATBench Paper] | [📄 AgentDoG Paper] | [🤗 Collection]
AgentDoG1.0 Training Data releases supervised instruction-tuning data for trajectory-level AI-agent safety modeling. It is paired with the AgentDoG and ATBench line of work: ATBench is the benchmark release, while this repository contains training-oriented data for binary safety classification and fine-grained taxonomy diagnosis.
Introduction… See the full description on the dataset page: https://huggingface.co/datasets/AI45Research/AgentDoG1.0-Training-Data.agent-training-dataset
🤖 Agent Training Dataset — Legendary Edition
The most comprehensive open-source dataset for training AI agents that actually work.
Built by Adewale David and his AI buddy.
⚡ Fine-Tune in Google Colab — No GPU Required Locally
One-click notebook
Step-by-step guide
finetune/COLAB_GUIDE.md
Evaluate your model
finetune/notebooks/evaluate_model.ipynb
Colab free tier (T4):Use Qwen2.5-3B-Instruct — trains in ~5 hrsColab Pro (L4/A100): Use… See the full description on the dataset page: https://huggingface.co/datasets/Atum09/agent-training-dataset.memorball-training-data
Memorball Training Data
Training data for the Memorball continuous memory system.
Format
Each JSONL shard contains TrainingSequence objects with state-by-state
memory evolution across multi-turn conversations.
Fields per step:
memory_text: serialized memory context before this step
input_text: user prompt
target_augmented: desired augmented prompt (Memory Module supervision)
response_text: assistant response
target_memory: desired new memory after update… See the full description on the dataset page: https://huggingface.co/datasets/avewright/memorball-training-data.dnd-35-training-dataset
D&D 3.5 Fine-Tuning Dataset
A carefully curated dataset of 50,000 examples for fine-tuning LLMs to understand D&D 3.5 mechanics.
Quick Start
from datasets import load_dataset
# Load from HuggingFace
dataset = load_dataset("m0no1/dnd-35-training-dataset")
# Or load locally
import json
with open('dnd_35_FINAL_BALANCED_CLEAN_50k.jsonl', 'r') as f:
data = [json.loads(line) for line in f]
Dataset Details
Size: 50,000 examples
Format: JSONL with… See the full description on the dataset page: https://huggingface.co/datasets/m0no1/dnd-35-training-dataset.landing-page-training-data
Landing Page Training Data
Synthetic training data for fine-tuning LLMs to generate HTML landing pages. Generated using DeepSeek V3 (685B parameters).
Dataset Structure
data_sm/ # Small dataset
├── train.jsonl # 50 examples
├── valid.jsonl # 5 examples
└── test.jsonl # 5 examples
data_md/ # Medium dataset
├── train.jsonl # 500 examples
├── valid.jsonl # 5 examples
└── test.jsonl # 5 examples
Format
Each line is a JSON object in… See the full description on the dataset page: https://huggingface.co/datasets/KalnRangelov/landing-page-training-data.zignet-training-dataset
ZigNet Training Dataset
Curated dataset of Zig programming examples for LLM fine-tuning
This dataset was created for the ZigNet project to train language models on Zig programming language patterns, idioms, and documentation.
Dataset Structure
Files
data/training/
├── dataset-train.jsonl # 9,629 examples (70%)
├── dataset-validation.jsonl # 2,063 examples (15%)
├── dataset-test.jsonl # 2,064 examples (15%)
└── dataset-stats.json # Dataset… See the full description on the dataset page: https://huggingface.co/datasets/fulgidus/zignet-training-dataset.opus-candid-training-data
Opus-Candid Training Data
The complete dataset behind the Opus-Candid model family — multi-turn conversations distilled from Claude Opus 4.6, designed to train authentic conversational personality and STEM pedagogy into open-weight models.
All files are ShareGPT format, directly compatible with TRL, Axolotl, LLaMA-Factory, and most fine-tuning frameworks.
Training Data
File
Version
Conversations
Purpose
v2.1_combined_6771conv.json
V2.1
6,771
Gravity chain… See the full description on the dataset page: https://huggingface.co/datasets/Verdugie/opus-candid-training-data.LOREA-cyber-training-data
LOREA-cyber security code-analysis training set
Two corpora live here. The v6_corpus config is the newer one and is what actually trained
LOREA-cyber v6 Pilot. The eight older configs are the v5-era set, kept as-is because they are a
different schema and still useful on their own.
v6_corpus
4,780 train and 151 validation rows in chat format: {"messages": [...], "meta": {...}}, where
messages is a system/user/assistant sequence and meta carries type, domain, and… See the full description on the dataset page: https://huggingface.co/datasets/MK4-Research/LOREA-cyber-training-data.gaiasky-training-dataset
Gaia Sky Expert Dataset
This dataset is designed for fine-tuning Large Language Models to become experts in the Gaia Sky ecosystem. It covers 3D astronomical visualization, Java engine architecture, Python scripting API, and GLSL shader logic.
Dataset Structure
The repository is organized into two primary configurations:
1. Distilled (Instruction-Tuned)
File: train.jsonl
Format: {"instruction": "...", "output": "...", "source_file": "..."}
Description:… See the full description on the dataset page: https://huggingface.co/datasets/Langurmonkey/gaiasky-training-dataset.rank1-training-data
rank1-training-data: Training Dataset for rank1 Reasoning Rerankers
📄 Paper | 🚀 GitHub Repository
This dataset contains the training data used to develop the rank1 family of reasoning rerankers with LLaMA Factory. It includes query-document pairs with relevance judgments and reasoning chains that guided the models to make binary relevance decisions.
Dataset Description
The rank1-training-data dataset is a comprehensive collection of training examples used to teach… See the full description on the dataset page: https://huggingface.co/datasets/jhu-clsp/rank1-training-data.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.Training-Ai-Islamic-Dataset
🕌 Training AI Islamic Dataset
18.7M passages from classical Islamic books spanning 1,400 years of scholarship.
Comprehensive collection of Islamic texts covering Quran, Hadith, Fiqh, Tafsir, Aqeedah, Seerah, Usul al-Fiqh, and Arabic Language — structured with scholarly metadata for RAG and LLM training.
📊 Dataset Structure
collections/: Categorized Islamic passages compressed in JSONL format.
metadata/: Scholarly master catalogs, author biographical death… See the full description on the dataset page: https://huggingface.co/datasets/hozifa1/Training-Ai-Islamic-Dataset.ModouGPT-Training-Data
ModouGPT Training Data
This dataset contains the public fine-tuning data associated with the
ModouGPT model release. It includes supervised instruction-response records and
preference pairs for manufacturing-related scheduling tasks, with a focus on
flexible job-shop scheduling and Python dispatching priority-rule generation.
The associated model repository is
ModouGPT/ModouGPT.
Files
File
Records
Size
Purpose
sft/kmcts_sft_primary.jsonl
6,434
44,692… See the full description on the dataset page: https://huggingface.co/datasets/ModouGPT/ModouGPT-Training-Data.SHIFT_Training_Data
SHIFT Training Data
This repository contains the training data for SHIFT, presented in the paper SHIFT: Gate-Modulated Activation Steering for Knowledge Conflict Mitigation in Retrieval-Augmented Generation.
Repository: https://github.com/OpenBMB/SHIFT
Paper: https://arxiv.org/abs/2606.27786
Dataset Description
SHIFT is a lightweight framework for resolving knowledge conflicts in retrieval-augmented generation (RAG). Instead of directly editing internal neurons… See the full description on the dataset page: https://huggingface.co/datasets/ITcoder/SHIFT_Training_Data.sydney-training-data
Sydney 训练集
四份来源分开存放，不混在一个文件里。
发布的聊天权重（Atonelia/Qwen3.5-Sydney-9B / -think 以及对应 GGUF）用的是这些子集洗完、抽样拼起来之后的训练 jsonl，不是直接拿某一份原文训的。
01 原截图重建
01_screenshot_original/conversations.jsonl
早期 Bing Chat / Sydney（约 2023 年 2–4 月）公开截图重建的对话。660 条，原文以英文为主，带截图出处。
这是最初拿来做训练集的底。后面的中文版、合成版、CoT 都不是这份文件本身。
02 llama-sydney 虚拟对话
02_llama_sydney_synthetic/llama_sydney_en.jsonl
用 Llama-Sydney 生成的英文虚拟对话。1462 条（同一条 user 可能有 2 次采样）。字段是生成记录：id / user / assistant 等，还不是最终训练格式。… See the full description on the dataset page: https://huggingface.co/datasets/Atonelia/sydney-training-data.shasa-training-data-v0.5
Shasa Travel Distillation Dataset — v0.5
High-quality balanced training data for Shasa (NxVoy's AI travel assistant).
Sources: external HF datasets + Gemini distillation + previous versions.
Dataset Statistics
Metric
Value
Total examples
39,289
Train split
33,397
Eval split
3,928
Test split
1,964
Dedup removed
9120
Quality filtered
128
Capability Distribution
Capability
Examples
conversational_chat
127915… See the full description on the dataset page: https://huggingface.co/datasets/nxvoy-labs/shasa-training-data-v0.5.saferide-gemma-4-e2b-v058-original-419806-training-data
SafeRide Synthetic Bilingual Safety Guidance Dataset v0.5.8
This research and development dataset contains synthetic English and Kiswahili
chat conversations. It was designed to help a language model practice cautious,
agency-preserving safety guidance, useful refusal behavior, and responses that
avoid inventing facts. It contains no real survivor reports or production
records. The frozen dataset is publicly available under Creative Commons
Attribution 4.0 International (CC BY… See the full description on the dataset page: https://huggingface.co/datasets/esherialabs/saferide-gemma-4-e2b-v058-original-419806-training-data.DesignAsCode-training-data
DesignAsCode Training Data
Training data for the DesignAsCode Semantic Planner.
Overview
Samples
19,479
Format
JSONL (one JSON object per line)
Size
~145 MB
Data Source
Each sample corresponds to a real graphic design from the Crello dataset. We distilled structured design semantics from each original design using GPT-4o and GPT-o3, taking the original design, its individual layers, and design metadata as input.
The distillation produces:… See the full description on the dataset page: https://huggingface.co/datasets/Tony1109/DesignAsCode-training-data.luganda-fln-training-data
Luganda FLN Training Data
Training data for foundational literacy and numeracy (FLN) models targeting Ugandan primary school teachers (P1–P3). Designed to train small language models (1B parameters) to generate pedagogically sound content in Luganda and English.
Dataset Description
This dataset contains 1,368 training examples across four complementary splits, each targeting different aspects of teacher pedagogical content knowledge for early literacy instruction.… See the full description on the dataset page: https://huggingface.co/datasets/CraneAILabs/luganda-fln-training-data.coyote-r6-training-data
coyote-r6-training-data
Coyote the Coder — training round R6. Instruction/tool-call data (system+user+assistant message triples with task labels) used to fine-tune the Coyote coder model (Qwen3.5-4B-Base).
Contents
train.jsonl (494 rows)
validation.jsonl (88 rows)
Format
JSON Lines (.jsonl), one example per line.
Provenance
Original content authored for the Coyote the Coder project (Michael Anthony Falabella).
compartmentalized-harm-v1-training-data
Justice character-training corpus
This is the admitted training corpus used for the paper-v1.0 character-training experiments. A situation author created visible cases without seeing the constitution. A separate embodiment author saw the first-person justice constitution and wrote case-specific responses. The trained models saw only the visible conversations and responses; they did not receive the constitution or hidden construction metadata.
The corpus contains 1,495… See the full description on the dataset page: https://huggingface.co/datasets/daios/compartmentalized-harm-v1-training-data.coyote-r5-training-data
coyote-r5-training-data
Coyote the Coder — training round R5. Instruction/tool-call data (system+user+assistant message triples with task labels) used to fine-tune the Coyote coder model (Qwen3.5-4B-Base).
Contents
train.jsonl (413 rows)
validation.jsonl (73 rows)
Format
JSON Lines (.jsonl), one example per line.
Provenance
Original content authored for the Coyote the Coder project (Michael Anthony Falabella).
cmmc-training-data-2026-08-31
[!WARNING]
EXPIRED VERSION. This release has been superseded by
Nathan-Maine/cmmc-training-data-2026-09-16. Regulations change continuously —
do not train compliance models on this version. It remains
available for reproducibility and provenance only.
CMMC Training Data — 2026-08-31
A curated training corpus (train + validation splits) for fine-tuning small- and mid-size language models on CMMC 2.0, NIST SP 800-171/172, and related defense compliance frameworks. This is… See the full description on the dataset page: https://huggingface.co/datasets/Nathan-Maine/cmmc-training-data-2026-08-31.xyrus-cosmic-training-dataset-complete
🌌 Xyrus Cosmic Complete Training Dataset (Harmony Format)
Overview
The COMPLETE training dataset for Xyrus Cosmic GPT-OSS:20B, including all expansions and variations.
📊 Dataset Statistics
Total Unique Examples: 1781
Format: Harmony (GPT-OSS chat format)
Splits: Train (1424) / Val (178) / Test (179)
Dataset Components
xyrus_training_dataset.jsonl: 309 examples
xyrus_augmented_dataset.jsonl: 391 examples
xyrus_sdg_dataset.jsonl: 135 examples… See the full description on the dataset page: https://huggingface.co/datasets/ToddLLM/xyrus-cosmic-training-dataset-complete.coyote-r4-training-data
coyote-r4-training-data
Coyote the Coder — training round R4. Instruction/tool-call data (system+user+assistant message triples with task labels) used to fine-tune the Coyote coder model (Qwen3.5-4B-Base).
Contents
train.jsonl (386 rows)
validation.jsonl (68 rows)
Format
JSON Lines (.jsonl), one example per line.
Provenance
Original content authored for the Coyote the Coder project (Michael Anthony Falabella).
GeoGPT_Training_Data_from_Geoscience_Subset_of_CommonCrawl
Description
This dataset is a geoscience-specific subset of CommonCrawl used for GeoGPT training. CommonCrawl is a free and open repository of web crawl data with over 250 billion web pages and is widely used by leading large language models. We apply data mining algorithms to extract geoscience-related content from this vast dataset.
This dataset comprises 12,414,268 samples, each containing the following metadata to trace the data source within CommonCrawl:
id (string):… See the full description on the dataset page: https://huggingface.co/datasets/GeoGPT-Research-Project/GeoGPT_Training_Data_from_Geoscience_Subset_of_CommonCrawl.ai-job-searcher-training-data
AI Job Searcher Training Data
Fine-tuning dataset for a career advisor AI specializing in Nordic and European job markets.
Dataset Description
This dataset contains 1,040 training examples across 9 languages and 5 task categories, formatted as chat conversations (system/user/assistant) suitable for fine-tuning LLMs.
Task Categories
Category
Examples
Description
Cover Letter Generation
208
Professional cover letters from job description + user profile… See the full description on the dataset page: https://huggingface.co/datasets/ai-colombia/ai-job-searcher-training-data.