datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
dnd5e-srd-qa
D&D 5.2.1 SRD RAG Evaluation Dataset
A high-quality Question-Answering (QA) dataset built by the Datapizza AI Lab from the Dungeons & Dragons 5th Edition System Reference Document (SRD) version 5.2.1, designed to evaluate Retrieval Augmented Generation (RAG) systems.
Dataset Summary
This dataset contains 56 question-answer pairs across two difficulty tiers (Easy and Medium), each designed to test different aspects of RAG system capabilities. The dataset is built from 20… See the full description on the dataset page: https://huggingface.co/datasets/datapizza-ai-lab/dnd5e-srd-qa.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.dnd-mechanics-dataset
D&D 3.5 Mechanics Dataset v4 (Math & Non-Math Fixes)
This dataset contains 40,365 D&D 3.5 mechanics question-answer pairs for fine-tuning language models to understand and apply D&D rules as a Dungeon Master.
Dataset Description
Overview
This dataset is designed to train language models to act as knowledgeable D&D 3.5 Dungeon Masters, capable of:
Resolving combat actions with proper dice rolls
Applying spell effects and restrictions
Handling skill checks and… See the full description on the dataset page: https://huggingface.co/datasets/m0no1/dnd-mechanics-dataset.dnd_dm_v3
DnD Dungeon Master Training Data V3
This dataset contains training data for AI Dungeon Masters in Dungeons & Dragons gameplay.
Dataset Description
This dataset consists of 1,633 examples of Dungeon Master interactions from DnD game sessions. The data is formatted for training language models to act as game masters, including:
Narrative storytelling: Describing scenes, environments, and events
NPC management: Controlling non-player characters and monsters
Game state… See the full description on the dataset page: https://huggingface.co/datasets/autoprogrammer/dnd_dm_v3.dnd-training-sharegpt-dm
DnD Dungeon Master Training Dataset
ShareGPT格式的D&D地下城主（DM）训练数据集，用于LLamaFactory微调。
数据集概览
记录数: 1,151条
文件大小: 43.31 MB
角色: DM (Dungeon Master / 地下城主)
工具数: 40个
DM角色说明
DM是D&D游戏的主持人，负责：
⚔️ 管理战斗流程（先攻、回合顺序）
🎲 裁定攻击和伤害
🧙 处理法术效果
🎭 扮演NPC和怪物
📊 跟踪所有角色状态
Tools (40个)
DM拥有完整的游戏管理工具：
战斗管理 (8个)
roll_initiative - 先攻投骰
roll_attack - 攻击检定
roll_dmg - 伤害投骰
roll_spell_attack - 法术攻击
roll_save - 豁免检定
roll_skill_check - 技能检定
opportunity_attack - 借机攻击… See the full description on the dataset page: https://huggingface.co/datasets/autoprogrammer/dnd-training-sharegpt-dm.dnd_characters_backstoriesThis dataset is made from this repo here
and it contains 2322 character bios to be used
dnd-training-qwen3
DnD Combat Training Dataset (Qwen3 Format)
训练AI模型扮演龙与地下城（D&D）游戏中的DM（地下城主）和玩家角色。
数据集概览
总记录数: 2,268条
DM样本: 1,151条 (50.7%)
Player样本: 1,117条 (49.3%)
文件大小: 59MB
格式: Qwen3-32B兼容格式，支持Function Calling
数据来源
本数据集来自DnD-Agents项目的teacher_io数据，包含：
场景: ambush（伏击）、cave（洞穴）、kennel（狗舍）
方法: bbps, cfrw, dmrw
难度: high, mid, low
共108个源文件，每个文件21条对话。
数据格式
每条记录包含完整的对话历史，符合Qwen3 chat模板格式：
{
"messages": [
{
"role": "system",
"content": "游戏规则和角色说明..."… See the full description on the dataset page: https://huggingface.co/datasets/autoprogrammer/dnd-training-qwen3.dnd-training-sharegpt
DnD Combat Training Dataset - ShareGPT Format
ShareGPT格式的D&D战斗训练数据集，用于LLamaFactory微调。
数据集概览
本数据集包含两个子集：
文件
角色
记录数
大小
Tools数量
dnd_training_data_sharegpt_dm.jsonl
DM (地下城主)
1,151
43.31 MB
40个
dnd_training_data_sharegpt_player.jsonl
Player (玩家)
1,117
18.74 MB
17个
总计
-
2,268
62.05 MB
-
格式说明
标准ShareGPT对话格式，兼容LLamaFactory：
{
"conversations": [
{
"from": "system",
"value": "游戏规则说明 + Tools定义（XML格式）"
},
{
"from":… See the full description on the dataset page: https://huggingface.co/datasets/autoprogrammer/dnd-training-sharegpt.dnd-mechanics-training-v2
D&D Mechanics Training Dataset v2
This dataset trains Qwen 2.5 7B to act as a D&D 3.5 game master, handling:
Player action resolution
Dice roll mechanics
Character queries
Rules arbitration
System orchestration
Dataset Structure
Format
Alpaca-style format compatible with Axolotl:
{
"instruction": "I attack the orc with my sword",
"input": "",
"output": "Roll attack: 1d20+5 vs orc's AC. Hit: 1d8+3 slashing damage."
}
Statistics
Training… See the full description on the dataset page: https://huggingface.co/datasets/m0no1/dnd-mechanics-training-v2.dnd-training-sharegpt-player
DnD Player Training Dataset
ShareGPT格式的D&D玩家角色训练数据集，用于LLamaFactory微调。
数据集概览
记录数: 1,117条
文件大小: 18.74 MB
角色: Player (玩家)
工具数: 17个
Player角色说明
Player是游戏参与者，负责：
🎭 扮演自己的角色（战士、法师、盗贼等）
🎯 决定行动和策略
🗣️ 与队友协作沟通
🔍 查询游戏状态和信息
⚔️ 在DM指导下进行战斗
Tools (17个)
Player拥有信息查询和角色控制工具：
状态查询 (9个)
check_hp - 查看生命值
check_ac - 查看护甲等级
check_buffs - 查看增益状态
check_concentration - 查看专注法术
check_resist - 查看抗性/免疫/脆弱
check_resources - 查看行动/法术位资源
check_class - 查看职业… See the full description on the dataset page: https://huggingface.co/datasets/autoprogrammer/dnd-training-sharegpt-player.dnd_player_v3
DnD Player Character Training Data V3
This dataset contains training data for AI player characters in Dungeons & Dragons gameplay.
Dataset Description
This dataset consists of 609 examples of player character interactions from DnD game sessions. The data is formatted for training language models to act as strategic game players, including:
Strategic decision-making: Choosing optimal actions in combat and exploration
Role-playing: Speaking and acting in character… See the full description on the dataset page: https://huggingface.co/datasets/autoprogrammer/dnd_player_v3.dnd5esrd_spellsA dataset of spells (with metadata) from Dungeons & Dragons 5th Edition SRD. Built with data from here.
Made just for fun! As such might include HTML parsing artifacts. Tread carefully.