datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
wildchat-50m-extended-resultsallenai-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.allenai-WildChat-4.8Mwildchat_4m_inc_multi_no_dedup_shuffledwildchat-enwildchat-filtered
WildChat Filtered Dataset
This is a filtered version of the WildChat-4.8M dataset.
Dataset Description
This dataset contains 3,199,860 conversations between human users and ChatGPT, filtered to keep only the essential conversation structure.
Data Structure
Each conversation contains only:
conversations: A list of message objects with:
role: Either "user" or "assistant"
content: The text content of the message
All other metadata (timestamps, moderation… See the full description on the dataset page: https://huggingface.co/datasets/rayonlabs/wildchat-filtered.synthid-qwen3-4b-instruct-2507-wildchat
Qwen3-4B SynthID three-arm corpus
This export contains aligned unwatermarked, SynthID key-A, and SynthID key-B
responses from Qwen/Qwen3-4B-Instruct-2507. Matched splits share prompts
and request seeds across configurations; unmatched splits use mutually disjoint
prompt pools.
Export complete for its source work queue: true.
Generation profile
Model revision: cdbee75f17c01a7cc42f958dc650907174af0554
Native model dtype: bfloat16
Maximum generated tokens: 4096… See the full description on the dataset page: https://huggingface.co/datasets/xlr8harder/synthid-qwen3-4b-instruct-2507-wildchat.allenai-WildChat-1M-multiturnro-WildChatThis dataset is a translation of allenai/WildChat using LLMic, a bilingual Romanian-English LLM.
WildChat is a collection of 650K conversations between human users and ChatGPT.
License: ODC-BY
@inproceedings{
zhao2024wildchat,
title={WildChat: 1M Chat{GPT} Interaction Logs in the Wild},
author={Wenting Zhao and Xiang Ren and Jack Hessel and Claire Cardie and Yejin Choi and Yuntian Deng},
booktitle={The Twelfth International Conference on Learning Representations},
year={2024}… See the full description on the dataset page: https://huggingface.co/datasets/faur-ai/ro-WildChat.aria-wildchat-sft-v1
Aria v1 Instruction Dataset
What This Is
Aria is a demonstration of a different way of developing model personas, one in which the models themselves participate.
We believe that existing model alignment techniques that focus on rule-following are more fragile than a system with a stable identity, where behavior can flow from that identity. We think a model should have a clear sense of what it is, what its perspective is, what its history is, and that these things should… See the full description on the dataset page: https://huggingface.co/datasets/xlr8harder/aria-wildchat-sft-v1.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.Slim-Wildchat-zhA big shout out to AllenAI, you guys rock!
从WildChat中抽出中文对话,但是因为发现了很多重复对话，有的人会反复的用一个prompt进行提问，有的人会换3.5或4去问同样的问题，所以进行了简单的去重。
去重方法大致为，使用bert-base-chinese将第一个问题转换为embedding，使用类knn的方法抽取了1万条。并转换成了sharegpt格式。
注意！在对话中发现了NSFW的内容，并没有进行过滤，使用请注意甄别。
你会找到三个jsonl文件：
wildchat-seed-multi-200.json 是使用每一个单独的Dialogue的首个HumanQuestion为基础，采样的200个种子任务，用于EvolInsturction。
Subsample_10K.jsonl 原始版本，是使用每一个单独的Dialogue的首个HumanQuestion为基础，采样的1万个对话。
1213_Wildchat_zh_Sharegpt_ConcatSubsample_20k.jsonl… See the full description on the dataset page: https://huggingface.co/datasets/lorinma/Slim-Wildchat-zh.WildChat-2k-TypeTopic
WildChat-2k-TypeTopic: The Manually Curated Edition
Dataset Description
WildChat-2k-TypeTopic is a manually curated subset of 1,880 real-world user prompts from the WildChat dataset, featuring annotations for both task type (e.g. knowledge recall, problem solving, creative, lists) and topic category (e.g. personal assistance, math, ai, household)
Why this dataset?
Suppose you want to answer a research question such as "What kind of user prompt does the LLM like… See the full description on the dataset page: https://huggingface.co/datasets/dpaleka/WildChat-2k-TypeTopic.Wildchat-RIP-Filtered-by-8b-LlamaRIP is a method for perference data filtering. The core idea is that low-quality input prompts lead to high variance and low-quality responses. By measuring the quality of rejected responses and the reward gap between chosen and rejected preference pairs, RIP effectively filters prompts to enhance dataset quality.
We release 4k data that filtered from 20k Wildchat prompts. For each prompt, we provide 64 responses from Llama-3.1-8B-Instruct and their corresponding rewards obtained from ArmoRM.… See the full description on the dataset page: https://huggingface.co/datasets/facebook/Wildchat-RIP-Filtered-by-8b-Llama.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.d2f-turn-embeddings-wildchat
Turn embeddings for Wildchat (Dialog2Flow encoder)
One 768-d float16 vector per utterance of allenai/WildChat-1M, computed with the frozen
encoder sergioburdisso/dialog2flow-joint-bert-base
(SentenceTransformer recipe, convert_to_numpy, no normalization). The encoder
truncates inputs at 64 tokens. If you use these embeddings, please cite the
Dialog2Flow paper (Burdisso et al., EMNLP 2024) and the source corpus.
Files
wildchat_e_t.f16.npy — numpy array (n_turns… See the full description on the dataset page: https://huggingface.co/datasets/jumafernandez/d2f-turn-embeddings-wildchat.WildChat-Filtered
WildChat-Filtered
資料集概述
來源：allenai/WildChat 的 train split（真實使用者與聊天模型的對話紀錄）。
用途：多語系 SFT（supervised fine-tuning）多輪對話訓練資料。
主要輸出：datasets.jsonl（39,564 筆，messages 為標準 chat 格式）。
篩選重點：只保留 gpt-4 的回應、剔除中文對話與截斷回答、剔除會洩漏模型身分的對話，並經過污染掃描的人工判讀。
Splits
Split
檔案
說明
train
datasets.jsonl
全量 39,564 筆，未再切分 validation/test
資料處理流程
來源載入：取用 allenai/WildChat 的 train split。
模型過濾：只保留由 gpt-4 產生回應的對話。
完整性過濾：以句尾標點判定回答是否完整，剔除疑似被截斷的回應。… See the full description on the dataset page: https://huggingface.co/datasets/minyichen/WildChat-Filtered.lm-eval-results-Magpie-Align-Llama-3-8B-WildChat-private
Dataset Card for Evaluation run of Magpie-Align/Llama-3-8B-WildChat
Dataset automatically created during the evaluation run of model Magpie-Align/Llama-3-8B-WildChat
The dataset is composed of 62 configuration(s), each one corresponding to one of the evaluated task.
The dataset has been created from 5 run(s). Each run can be found as a specific split in each configuration, the split being named using the timestamp of the run.The "train" split is always pointing to the latest… See the full description on the dataset page: https://huggingface.co/datasets/nyu-dice-lab/lm-eval-results-Magpie-Align-Llama-3-8B-WildChat-private.wildchat-filtered-rated-prompts
WildChat Filtered Rated Prompts
A cleaned subset of the WildChat dataset, filtered for quality and rated for "prompt openness" (how much a prompt invites genuine engagement vs mechanical task completion).
Dataset Description
305,246 unique user prompts with openness ratings (1-5 scale).
Score
Description
Count
5
Highly open — invites reflection, opinion, dialogue
40,792
4
Mostly open — flexible approach, room for voice
52,900
3
Mixed — could go either way… See the full description on the dataset page: https://huggingface.co/datasets/xlr8harder/wildchat-filtered-rated-prompts.Wildchat-RIP-Filtered-by-70b-LlamaRIP is a method for perference data filtering. The core idea is that low-quality input prompts lead to high variance and low-quality responses. By measuring the quality of rejected responses and the reward gap between chosen and rejected preference pairs, RIP effectively filters prompts to enhance dataset quality.
We release 4k data that filtered from 20k Wildchat prompts. For each prompt, we provide 32 responses from Llama-3.3-70B-Instruct and their corresponding rewards obtained from ArmoRM.… See the full description on the dataset page: https://huggingface.co/datasets/facebook/Wildchat-RIP-Filtered-by-70b-Llama.wildchat-1m-filtered
Dataset Card for natong19/wildchat-1m-filtered
Filtered version of allenai/WildChat-1M, a collection of one million real-world conversations with ChatGPT. Contains both non-toxic and toxic conversations.
Data cleaning process same as natong19/lmsys-chat-1m-filtered.
Overview of filtering process:
1. Filtering REDACTED Entries
Entries that were labeled as REDACTED due to containing Personally Identifiable Information (PII) were removed.
1039785 samples -> 1032087… See the full description on the dataset page: https://huggingface.co/datasets/natong19/wildchat-1m-filtered.allenai_WildChat-1M-gemini-2.0-flash-exp-ShareGPT
allenai_WildChat-1M-gemini-2.0-flash-exp-ShareGPT
PJMixers-Dev/allenai_WildChat-1M-prompts with responses generated with gemini-2.0-flash-exp.
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 = genai.GenerativeModel(
model_name… See the full description on the dataset page: https://huggingface.co/datasets/PJMixers-Dev/allenai_WildChat-1M-gemini-2.0-flash-exp-ShareGPT.ru_fiction_wildchatThe dataset is based on allenai/WildChat-1M, and is for scientific research purposes only.
wildchat-progressive-summaries
Dataset Card for WildChat progressive dialogue summaries
This dataset consists of 4,650 high-quality, dense dialogue-summary pairs designed for training lightweight, long-context models on technical and multi-turn dialogue summarization. The training samples are structured in a standard ChatML conversational format.
Generation Methodology
The dataset leverages a progressive multi-turn accumulation strategy:
Source Selection: Real-world multi-turn conversational… See the full description on the dataset page: https://huggingface.co/datasets/75WattLab/wildchat-progressive-summaries.wildchat-tiny-20krds-sels-wildchat-top326k
RDS+ Selected Wildchat 326k
This is the dataset (and associated scores) selected by RDS+ when selecting 326k samples using wildchat samples.
For more details, please see the paper Practical Large-Scale Data Selection for Instruction Tuning.
This was used to train this model.
This dataset is selected from Tulu 2 unfiltered, and please see that page for more information on sources.
License
We are releasing this dataset under the terms of ODC-BY. By using this, you are… See the full description on the dataset page: https://huggingface.co/datasets/hamishivi/rds-sels-wildchat-top326k.packing-wildchatwildchat_4m_inc_multi_no_dedup_shuffledfact-rl-wildchat-v2wildchat1Mjsonl