Dataset Detail

Unknown / mixed-size (kept for completeness; format consistent with original)

• CodeParrot - | python | 180GB Python files (<1MB each), 20M+ files. | -
• Alpaca‑CoT - | Multi-lingual | Instruction data with chain‑of‑thought traces. | odc-by
• stack-exchange-paired - | English | StackExchange Q&A pairs for preference modeling. | cc-by-sa-4.0
• LangChainDatasets - | English | Community datasets to evaluate chains & agents. | -
• ParlAI - | English | Dialog research platform with many tasks/datasets. | -
• GPTeacher - | English | Instruction datasets consolidated for general SFT. | -
• Wizard‑LM Chinese Evol - | Chinese | Chinese evol‑instruct corpus. | -
• ToolACE - | English | Multi‑tool calling SFT (functions, API JSON, tool plans). | -
• UltraFeedback (cleaned binarized) - | English | UltraFeedback preferences cleaned & binarized. | cc-by-nc-4.0
• glaive‑function‑calling‑v2 - | English | Function‑calling SFT dataset with tool schemas & arguments. | apache-2.0

• TheoremQA
• LIMA - by-nc-sa-4.0
• im-feeling-curious
• Puffin - turn examples; each response via GPT‑4. | apache-2.0
• cc_sbu_align - 3-clause
• qa_feedback - construct the ASQA data and collect human feedback for it. We name the resulting dataset as qa-feedback. | - |
• SLF5K - 2.0
• blended_skill_talk - |
• PKU‑SafeRLHF‑10K
• Dolly - dolly-15k is a corpus of more than 15,000 records generated by thousands of Databricks employees to enable large language models to exhibit the magical interactivity of ChatGPT. | CC 3.0 |
• WebGPT (comparisons)
• LongForm
• HC3
• Mol‑Instructions - by-4.0
• arxiv‑math‑instruct‑50k (ArtifactAI)
• Finance
• Vicuna Dataset - |
• InstructionTranslation - lingual | Translations were generated by M2M 12B and the output generations were limited at 512 tokens due to VRAM limit (40G). | MIT |
• Self-Instruct - |
• OASST1 - lingual | Human‑generated assistant conversations (35 languages). | apache-2.0
• HH-RLHF
• Guanaco Dataset - 3.0
• Tapir-Cleaned - tuning. | CC BY-NC 4.0 |
• WizardLM_evol_instruct_V2_196k - |
• LLaVA Visual Instruct 150K - by-nc-4.0
• ProsocialDialog
• COIG - 2.0
• SHP
• dromedary - Verbose-Clone is a synthetic dataset of 360k instructions and demonstrations. | cc-by-nc-4.0 |
• ultrachat - by-nc-4.0 |
• ign_clean_instruct_dataset_500k - instruction pairs with high quality responses. It was synthetically created from a subset of Ultrachat prompts. It does not contain any alignment focused responses or NSFW content. | apache-2.0 |
• ELI5 - language dataset of questions and answers gathered from three subreddits where users ask factual questions requiring paragraph-length or longer answers. | - |
• Instruct
• LaMini‑Instruction - by-nc-4.0
• BELLE
• OIG-43M Dataset - lingual | Together, LAION, and Ontocord.ai. | - |
• xP3 - lingual | 78,883,588 instructions from prompted datasets across 46 languages & 16 tasks. | -
• Alpaca-CoT Dataset - | Multi-lingual | Instruction Data Collection | ODC-By |
• stack-exchange-paired - | English | This dataset contains questions and answers from the Stack Overflow Data Dump for the purpose of preference model training. | cc-by-sa-4.0 |
• LangChainDatasets - | English | This is a community-drive dataset repository for datasets that can be used to evaluate LangChain chains and agents. | - |
• ParlAI - | English | 100+ popular datasets available all in one place, dialogue models, from open-domain chitchat, to task-oriented dialogue, to visual question answering. | - |
• silk-road/Wizard-LM-Chinese-instruct-evol - | chinese | Wizard-LM-Chinese | - |
• Chatbot Arena Conversations
• Anthropic_HH_Golden - tuned model of Anthropic, where harmful and unhelpful responses are freqently encountered. In this dataset, the positive responses are replaced by re-rewritten responses generated by GPT4. | |
• orca-chat - style dataset. The the process involves removing samples with very high similarity and also grouping instructions to form conversation. | |
• OpenOrca (full)
• MultiWOZ - | English | Multi-Domain Wizard-of-Oz dataset (MultiWOZ), a fully-labeled collection of human-human written conversations spanning over multiple domains and topics. | apache-2.0 |
• GSM‑IC
• ChatAlpaca - 2.0 license |
• Code Alpaca - |
• Traditional Chinese Alpaca Dataset - 2.0 license |
• Cabrita Dataset
• Japanese Alpaca Dataset
• Alpaca Dataset
• Alpaca Data Cleaned - |
• Alpaca GPT-4 Data (Chinese) - 4 using Chinese prompts translated from Alpaca by ChatGPT | - |
• InstructionWild
• Unnatural Instructions - ative and diverse instructions, collected with virtually no human labor. | MIT |
• GPT4All Dataset - lingual | Subset of LAION OIG, StackOverflow Question, BigSciense/p3 dataset. Answered by OpenAI API. | - |
• Natural Instructions - lingual | 5,040,134 instructions collected from diverse NLP tasks | - |
• Firefly - |
• GPTeacher - | English | A collection of modular datasets generated by GPT-4, General-Instruct - Roleplay-Instruct - Code-Instruct - and Toolformer | - |
• blended_skill_talk
• MOSS - 3.5-turbo | Apache-2.0, AGPL-3.0 licenses |
• TheoremQA
• WildGuardMix - annotator labels. | odc-by
• Berkeley Function Calling Leaderboard (BFCL) - calling eval covering parallel/multi-call scenarios across languages. | -
• QA-Feedback
• ChatAlpaca‑10K - 2.0
• Dolly‑15K - by-3.0
• CodeAlpaca‑20K
• HelpSteer2 - source helpfulness data for reward models and preference learning. | cc-by-4.0
• openapi-function-invocations‑25k - call traces. | mit
• LongForm
• HH‑RLHF
• RefGPT
• arxiv‑math‑instruct‑50k
• Traditional Chinese Alpaca - 2.0
• Cabrita Dataset
• Japanese Alpaca - by-nc-4.0; OpenAI terms
• Alpaca Dataset - by-nc-4.0; OpenAI terms
• Alpaca Data Cleaned
• Alpaca GPT‑4 Data
• Alpaca GPT‑4 Chinese
• xLAM Function Calling 60K - calling data for executable agents. | apache-2.0
• Dynosaur - 2.0
• WizardLM evol
• Vicuna Dataset
• InstructionTranslation - lingual | M2M‑12B translated instructions (≤512 tokens). | mit
• Instruct
• InstructionWild - only; OpenAI terms
• CAMEL Dataset
• TAPIR‑Cleaned - by-nc-4.0
• OASST2 (final) - lingual | Open Assistant Conversations Release 2 (train+val). | apache-2.0
• M2Lingual - lingual | Multilingual mixed‑modal (code+text) chat/instruct SFT. | -
• OpenR1‑Math‑220k - 2.0
• Unnatural Instructions
• WildJailbreak - by
• UltraChat - by-nc-4.0
• IGN Clean Instruct 500K - 2.0
• GPT4All - lingual | LAION OIG + StackOverflow + P3 prompts; OpenAI outputs. | -
• MOSS - 2.0 + agpl-3.0
• WildChat‑4.8M (nontoxic subset) - by
• smolTalk - 2.0
• Open‑PerfectBlend - 2.0
• The Tome
• NaturalReasoning - by-nc-4.0
• Infinity‑Instruct - lingual | 7.4M base + ~1.5M chat instruction data. | cc-by-sa-4.0
• BELLE‑10M - only; OpenAI terms
• Firefly
• OIG‑43M - lingual | LAION + Together + OntoCord composite instruction pool. | -

Unlock the Power of LLM: Explore These Datasets to Train Your Own ChatGPT!

• Dynosaur - tuning data curation. | Apache-2.0 license |