datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Creative_Writing_MultiturnUPDATE 2026: Stronger filtering using a very sophisticated filtering script and new data including a very small subset of https://huggingface.co/datasets/lemon07r/VellumK2T-Fiction-SFT-01 reasoning for thinking with a custom system prompt attached. This is suitable for both instruct non-thinking and thinking models, as I have added a system prompt for these few samples that use the tags <!think!> and </!think!> (without exclamation marks of course).
This is a dataset merge of many, many high… See the full description on the dataset page: https://huggingface.co/datasets/Dampfinchen/Creative_Writing_Multiturn.NousResearch-Hermes-3-Dataset-multiturn
Hermes 3 Multiturn
This is a filtered subset of NousResearch/Hermes-3-Dataset
containing only multiturn conversations with more than three messages.
Conversations with repetitive or trivial replies (for example, repeated "OK") have been excluded to improve quality.
multiturn_chat_0.8M
Multiturn Chat 0.8M
内容
包含约80万条由BELLE项目生成的用户与助手的多轮对话。
注意：此数据集是由ChatGPT产生的，未经过严格校验，内容可能包含错误。使用过程中请注意这一点。
instruction中包含多轮对话的上文内容，以Human:和Assistant:区分，output中包含当前助手角色的回答。
样例
{
"instruction":… See the full description on the dataset page: https://huggingface.co/datasets/BelleGroup/multiturn_chat_0.8M.Multi-turn_Long-context_Benchmark_for_LLMs
LoopServe: An Adaptive Dual-phase LLM Inference Acceleration System for Multi-Turn Dialogues
Arxiv: https://www.arxiv.org/abs/2507.13681
Huggingface: https://huggingface.co/papers/2507.13681
Introduction
LoopServe Multi-Turn Dialogue Benchmark is a comprehensive evaluation dataset comprising multiple diverse datasets designed to assess large language model performance in realistic conversational scenarios.
Unlike traditional benchmarks that place queries only at the end… See the full description on the dataset page: https://huggingface.co/datasets/TreeAILab/Multi-turn_Long-context_Benchmark_for_LLMs.Nemotron-RL-Instruction-Following-MultiTurnChat-v1
Dataset Description:
The MultiChallenge Dataset is a rigorous benchmark designed to improve large language models in complex multi-turn conversations by explicitly targeting inference memory, instruction retention, version editing, and self-coherence. It employs a unique "model breaking" methodology where tasks are tested against advanced models (Nemotron-Nano-V2 and Qwen3-235B-A22B-Thinking-2507) to expose failure modes. A sample is only accepted into the dataset if the task is… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-RL-Instruction-Following-MultiTurnChat-v1.Multi-Turn
TCOD: Exploring Temporal Curriculum in On-Policy Distillation for Multi-turn Autonomous Agents
Official codebase for TCOD, a temporal curriculum framework for on-policy distillation that stabilizes knowledge transfer from teacher to student agents in multi-turn interactive environments.
🔥 News
[2026-07] Our paper is accepted by COLM 2026!
[2026-06] ✍️ New blog post out: on-policy distillation pitfalls — sharing the lessons and pitfalls behind our… See the full description on the dataset page: https://huggingface.co/datasets/SeanWang0027/Multi-Turn.Creative-Writing-Multiturn-Cleaned-16kMultiTurnChat
CapX Scientific QA Dataset
The CapX Scientific QA Dataset is a comprehensive collection of data designed to assist in the development of AI-powered tools that support scientists across various disciplines. This dataset aims to bridge the gap between machine learning and the scientific community by providing reliable and transparent resources for training and evaluating question-answering systems.
Features
CapX Scientific QA Dataset
The CapX Scientific QA… See the full description on the dataset page: https://huggingface.co/datasets/Capx/MultiTurnChat.multiturn-chatmultiturnAll-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.multiturn_dataKnowledge-QA-MultiTurn-Dataset
Knowledge QA Multi-turn Dataset（知識質問データセット・マルチターン）
概要
本データセットは、Aratako/Synthetic-JP-Conversations-Magpie-Nemotron-4-10k から質問を抽出し、DeepSeek V3.2で整形・フォローアップ質問を生成、Kimi K2.5で回答を生成した 3ターンのマルチターン知識質問応答データセット です。Reasoning有効化により思考過程も最終データに含まれ、質問の難易度に応じてReasoning effortが動的に切り替わります。生成にはSDG-LOOMという合成データ生成パイプラインを用いました。（sdg-loom）
データの説明
項目
内容
件数
約3,000件
形式
JSONL（1行1JSON）
言語
日本語
ターン数
3ターン（質問3 + 回答3）
ソースデータセット… See the full description on the dataset page: https://huggingface.co/datasets/DataPilot/Knowledge-QA-MultiTurn-Dataset.0804calm3-logical-multiturn-pretrain
自動生成したテキスト
Calm3で自動生成したマルチターン会話のテキストです｡
一部の計算には東京工業大学のスーパーコンピュータTSUBAME4.0を利用しました｡
multi-turn-conversation-50000xTopics based on HuggingFaceTB/everyday-conversations-llama3.1-2k, expanded to 50000 examples. All converstations kept under 2000 tokens.
From source README:
##################
Everyday conversations for Smol LLMs finetunings
This dataset contains 2.2k multi-turn conversations generated by Llama-3.1-70B-Instruct. We ask the LLM to generate a simple multi-turn conversation, with 3-4 short exchanges, between a User and an AI Assistant about a certain topic.
The topics are chosen to be… See the full description on the dataset page: https://huggingface.co/datasets/kth8/multi-turn-conversation-50000x.allenai-WildChat-1M-multiturnMulti-Turn-Conversational-SFTCreated by: DataCreator AI
Multi-Domain Multi-Turn Chat Conversations Dataset
A synthetic conversational dataset designed for LLM supervised fine-tuning and chatbot training.
The dataset contains multi-turn dialogues across multiple everyday domains such as travel, banking, health, programming, and customer interactions. Conversations are structured in OpenAI chat fine-tuning format, making the dataset directly usable in modern fine-tuning pipelines.
Dataset Overview… See the full description on the dataset page: https://huggingface.co/datasets/DataCreatorAI/Multi-Turn-Conversational-SFT.multi-turn-chat-sft-50k
Multi-Turn Chat SFT (50K ShareGPT Format)
50,000 multi-turn conversations in ShareGPT format for supervised fine-tuning of chat models.
Format
Standard ShareGPT format — drop-in compatible with LLaMA-Factory, Axolotl, and Unsloth:
{
"conversations": [
{"from": "system", "value": "You are a helpful assistant."},
{"from": "human", "value": "Write a Python function to implement binary search."},
{"from": "gpt", "value": "Here's a clean… See the full description on the dataset page: https://huggingface.co/datasets/stindardlogic/multi-turn-chat-sft-50k.Multi_Turn_SWE_dataset
SWE-Agent Tool Trace Dataset
A structured software-engineering agent trajectory dataset derived from SWE-chat.
This dataset converts real-world coding-agent interaction traces into compact action-observation trajectories optimized for tool-use fine-tuning and autonomous software engineering agents.
The original SWE-chat dataset captures full coding sessions from real developers using coding agents such as Claude Code, Codex, Gemini CLI, Cursor, and others. :contentReference
This… See the full description on the dataset page: https://huggingface.co/datasets/suhas9545/Multi_Turn_SWE_dataset.glm-4.7-multiturn-CoT
glm-4.7-multiturn-CoT
Dataset Summary
glm-4.7-multiturn-CoT is a ShareGPT-style multi-turn reasoning distillation dataset generated with GLM-4.7 as the teacher model.
This release focuses on preserving multi-turn dialogue continuity while injecting explicit chain-of-thought style responses in assistant turns.
Key Features
Multi-turn conversation format (human / gpt)
Assistant responses stored as <think>...</think> + final answer
Resume-safe distillation… See the full description on the dataset page: https://huggingface.co/datasets/Jackrong/glm-4.7-multiturn-CoT.roleplay_multiturn_chat_1k_zh_v0.1
介绍
基于self-instruct生成的多轮对话roleplay数据，约1k条不同的人格数据和对话
存在问题：
1.基于模型自身生成，所以roleplay存在模型本身价值观融入情况，导致roleplay不够真实，不够准确。
关于我自己：
我是小雨的开发者，小雨是一个情感ai，人格ai，如果对小雨感兴趣的话欢迎支持一下，她目前在bilibili直播，目前我仍在不断的改进。未来，“小雨”的目标是成为一个
具有真正人类情感的多模态通用人工智能。
url：https://live.bilibili.com/27357528?broadcast_type=0&is_room_feed=1&spm_id_from=333.999.live_users_card.0.click&live_from=86001
注：
使用本数据集请注明来源
Introduction
This dataset consists of approximately 1,000 instances of… See the full description on the dataset page: https://huggingface.co/datasets/Minami-su/roleplay_multiturn_chat_1k_zh_v0.1.Multi-Turn-Instruct
Multi-Turn-Instruct Dataset
Dataset introduced in paper "Can Language Models Follow Multiple Turns of Entangled Instructions?"
📌 Overview
This repository contains the dataset, evaluation code, and benchmarks for the Multi-Turn-Instructdataset introduced in:
Can Language Models Follow Multiple Turns of Entangled Instructions?Chi Han, Xin Liu, Haodong Wang, Shiyang Li, Jingfeng Yang, Haoming Jiang, Zhengyang Wang, Qingyu Yin, Liang Qiu, Changlong Yu, Yifan Gao, Zheng… See the full description on the dataset page: https://huggingface.co/datasets/Glaciohound/Multi-Turn-Instruct.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-1999-2025.0723-calm3-22b-random-genre-inst-sft-multiturn-clean-tsub
自動生成Q&A
ランダムなジャンルについて､OpenCalm3-22bで生成したQ&Aです｡
一部の計算には東京工業大学のスーパーコンピュータTSUBAME4.0を利用しました｡
データ
クリーニングはしていません｡おかしなtextが一定数､含まれます
multi-turn-aware-quantization-llama-3.3-rp-testI added role headers and tokens for each turn in the LLaMA 3 Instruct format. The purpose is to test whether formatted multi-turn data can improve multi-turn performance after quantization.
Math-Multiturn-1K-ShareGPTAll samples were created with this script, no GPT, just python.
thai-qa-multiturn-answer-dataset
Thai QA Multiturns Answer Synthesis Dataset
Rows: 11,992 rows (Cleaned)
Generated by Kobkrit Viriyayudhakorn (kobkrit@iapp.co.th)
Examples
{"instruction": "[{\"human\": \"หวัดดี มีเรื่องอยากสอบถามเกี่ยวกับวิวัฒนาการของมนุษย์\"}]", "output": "สวัสดีค่ะ ยินดีเลยค่ะ มีคำถามอะไร ถามได้่เลยนะคะ"}
{"instruction": "[{\"human\": \"หวัดดี มีเรื่องอยากสอบถามเกี่ยวกับวิวัฒนาการของมนุษย์\"}, {\"assistant\": \"สวัสดีค่ะ ยินดีเลยค่ะ มีคำถามอะไร ถามได้่เลยนะคะ\"}, {\"human\":… See the full description on the dataset page: https://huggingface.co/datasets/openthaigpt/thai-qa-multiturn-answer-dataset.HuatuoGPT_sft_data_v1_multiturn
describe
process FreedomIntelligence/HuatuoGPT-sft-data-v1 to multiturn format
{
"instruction": "听起来很不错。人工智能可能在哪些方面面临挑战呢？",
"input": "",
"output": "人工智能面临的挑战包括数据隐私、安全和道德方面的问题，以及影响就业机会的自动化等问题。",
"history": [
["你好，你能帮我解答一个问题吗？", "当然，请问有什么问题？"],
["我想了解人工智能的未来发展方向，你有什么想法吗？", "人工智能在未来的发展方向可能包括更强大的机器学习算法，更先进的自然语言处理技术，以及更加智能的机器人。"]
]
}
which can be used at LLaMA-Efficient-Tuning
example
multi-turn-med-dialog
data summary
Multi-Turn Medical diaglogue of chinese & english
samples
zh: 1.5k
{
"dialog_id": 3,
"thread": [
{
"turn": 0,
"text": "眼睛白内障内膜薄怎么办（女，82岁）",
"role": "patient"
},
{
"turn": 1,
"text": "您好，您为什么要咨询内膜呢？",
"role": "doctor"
},
{
"turn": 2,
"text": "是怕晶状体囊膜太薄放不了人工晶体吗？",
"role": "doctor"
},
{
"turn": 3,
"text": "今天检验白内障，眼内膜只有500，医生说不能做手术。",
"role":… See the full description on the dataset page: https://huggingface.co/datasets/zirui3/multi-turn-med-dialog.Math-Multiturn-100K-ShareGPTAll samples were created with this script, no GPT, just python.