instruction stringlengths 44 183 | input stringlengths 212 978 | output stringlengths 234 734 | source stringclasses 1
value | original_id stringlengths 8 45 |
|---|---|---|---|---|
Given a raw dataset listing, identify whether it is suitable for supervised fine-tuning (SFT) of LLMs and summarize why. | Dataset: prompts.chat
Source: huggingface
ID: fka/prompts.chat
Description: a.k.a. Awesome ChatGPT Prompts. This is a dataset mirror of prompts.chat, a social platform for AI prompts where users share, discover, and collect AI prompts.
Tags: task_categories:question-answering, task_categories:text-generation, license:c... | This dataset is only partially suitable for SFT. It appears to be a prompt collection rather than a corpus of prompt-response training pairs. It can be useful for instruction template mining, prompt engineering research, or synthetic data generation, but by itself it likely lacks high-quality assistant completions need... | huggingface | fka/prompts.chat |
Given a raw dataset listing, identify whether it is suitable for supervised fine-tuning (SFT) of LLMs and summarize why. | Dataset: fineweb
Source: huggingface
ID: HuggingFaceFW/fineweb
Description: FineWeb consists of more than 18.5T tokens of cleaned and deduplicated English web data from CommonCrawl, optimized for LLM performance.
Tags: task_categories:text-generation, language:en, license:odc-by, size_categories:10B<n<100B, modality:ta... | This dataset is not ideal for supervised fine-tuning. FineWeb is a massive pretraining corpus of cleaned web text, which is excellent for continued pretraining or foundation model training, but it does not primarily provide instruction-response pairs or curated assistant demonstrations. Its main value is scale and text... | huggingface | HuggingFaceFW/fineweb |
Given a raw dataset listing, identify whether it is suitable for supervised fine-tuning (SFT) of LLMs and summarize why. | Dataset: hh-rlhf
Source: huggingface
ID: Anthropic/hh-rlhf
Description: This repository provides human preference data about helpfulness and harmlessness for training preference or reward models for RLHF. The dataset card explicitly notes these data are not meant for supervised training of dialogue agents.
Tags: licens... | This dataset is generally not suitable for direct SFT. It is designed for preference modeling and RLHF reward training rather than supervised dialogue training. The dataset is highly valuable for alignment research, especially for learning preference or reward models around helpfulness and harmlessness, but the card ex... | huggingface | Anthropic/hh-rlhf |
Given a raw dataset listing, identify whether it is suitable for supervised fine-tuning (SFT) of LLMs and summarize why. | Dataset: OpenOrca
Source: huggingface
ID: Open-Orca/OpenOrca
Description: The OpenOrca dataset is a rich collection of augmented FLAN data aligned with distributions outlined in the Orca paper and has been used to generate high-performing model checkpoints.
Tags: task_categories:text-classification, token-classificatio... | This dataset is highly suitable for SFT. OpenOrca is an instruction-style dataset derived from FLAN-style augmentation and explicitly intended to support strong model training. It likely contains prompt-completion examples across many task types, making it useful for general-purpose instruction tuning. Additional stren... | huggingface | Open-Orca/OpenOrca |
Given a raw dataset listing, identify whether it is suitable for supervised fine-tuning (SFT) of LLMs and summarize why. | Dataset: oasst1
Source: huggingface
ID: OpenAssistant/oasst1
Description: OpenAssistant Conversations Dataset (OASST1) is a human-generated, human-annotated assistant-style conversation corpus with 161,443 messages in 35 languages and extensive quality ratings across annotated conversation trees.
Tags: many languages i... | This dataset is highly suitable for SFT. OASST1 contains human-generated assistant conversations and annotations, which makes it directly relevant for training conversational LLMs. Its multilingual coverage is a major advantage for cross-lingual or multilingual assistants, and the quality ratings can also support filte... | huggingface | OpenAssistant/oasst1 |
Summarize the dataset in a concise way for supervised fine-tuning. | Name: gsm8k
Description: Dataset Card for GSM8K. GSM8K (Grade School Math 8K) is a dataset of 8.5K high quality linguistically diverse grade school math word problems created to support question answering on basic mathematical problems requiring multi-step reasoning. Problems typically take 2 to 8 steps and involve ele... | GSM8K is an English grade-school math word problem dataset designed for multi-step reasoning and question answering. It contains about 8.5K high-quality problems that usually require 2 to 8 arithmetic steps to solve. It is commonly used as a benchmark for mathematical reasoning. | huggingface | openai/gsm8k |
Summarize the dataset in a concise way for supervised fine-tuning. | Name: wikipedia
Description: Dataset Card for Wikimedia Wikipedia. Wikipedia dataset containing cleaned articles of all languages. Built from Wikipedia dumps with one subset per language, each containing a single train split. Each example contains the content of one full Wikipedia article with cleaning to strip markdow... | Wikimedia Wikipedia is a large multilingual corpus of cleaned Wikipedia articles derived from official dumps. It provides one subset per language, with each record containing the cleaned text of a full article. The dataset is useful for language modeling, masked language modeling, and general text generation tasks. | huggingface | wikimedia/wikipedia |
Summarize the dataset or resource in a concise way for supervised fine-tuning. | Name: EasyNegative
Description: Negative Embedding trained with Counterfeit. Intended for use in the stable-diffusion-webui embeddings folder. It may also be used with other models, though effectiveness is uncertain.
Tags: license:other, format:imagefolder, modality:image
Downloads: 35800
Likes: 1174 | EasyNegative is a negative embedding resource for image generation workflows, especially Stable Diffusion web UI setups. It was trained with Counterfeit and is intended to be placed in the embeddings folder to influence generation behavior. It is not a standard text dataset, but rather an image-related model asset. | huggingface | gsdf/EasyNegative |
Summarize the dataset in a concise way for supervised fine-tuning. | Name: RedPajama-Data-1T
Description: RedPajama is a clean-room, fully open-source implementation of the LLaMa dataset.
Tags: task_categories:text-generation, language:en, modality:text
Downloads: 2142
Likes: 1149 | RedPajama-Data-1T is a large-scale English text dataset created as an open-source, clean-room reproduction of the data used for LLaMA-style model training. It is intended for text generation and foundation model pretraining use cases. | huggingface | togethercomputer/RedPajama-Data-1T |
Summarize the dataset in a concise way for supervised fine-tuning. | Name: medical-o1-reasoning-SFT
Description: Medical reasoning dataset for SFT built on medical verifiable problems and LLM-generated reasoning chains. News notes mention a medical-only SFT split, a mixed file with medical and general instruction data, and a distilled dataset from Deepseek-R1 for medical verifiable prob... | medical-o1-reasoning-SFT is a bilingual English-Chinese medical reasoning dataset for supervised fine-tuning. It focuses on medical verifiable problems and includes reasoning chains intended to help train models on structured medical question answering and reasoning tasks. The dataset is especially relevant for medical... | huggingface | FreedomIntelligence/medical-o1-reasoning-SFT |
Summarize the dataset FineWeb-Edu for LLM pretraining use, including its scale, purpose, language, license, format, and notable metadata. | Source: huggingface
Dataset ID: HuggingFaceFW/fineweb-edu
Name: fineweb-edu
Description: FineWeb-Edu is a 1.3 trillion token educational web dataset filtered from FineWeb using an educational quality classifier. It is the 1.3T token version and is intended for high-quality educational language model training.
Tags: tas... | FineWeb-Edu is a large-scale English educational text dataset on Hugging Face designed for language model pretraining and text generation research. It contains 1.3 trillion tokens of educational web content filtered from the FineWeb corpus using an educational quality classifier. The dataset is distributed under the OD... | huggingface | HuggingFaceFW/fineweb-edu |
Provide a concise dataset description for Dolma, covering its purpose, token scale, language, license, and likely use case in LLM training. | Source: huggingface
Dataset ID: allenai/dolma
Name: dolma
Description: Dolma: an Open Corpus of Three Trillion Tokens for Language Model Pretraining Research
Tags: task_categories:text-generation, language:en, license:odc-by, size_categories:n>1T, arxiv:2402.00159, arxiv:2301.13688, region:us, language-modeling, casual... | Dolma is an open English-language corpus created for language model pretraining research. It contains three trillion tokens, making it a very large-scale resource suitable for training and studying large language models. The dataset is tagged for text generation, language modeling, causal LM, and LLM use cases, and is ... | huggingface | allenai/dolma |
Summarize The Stack dataset with emphasis on its content type, multilingual/code nature, licensing notes, format, and scale. | Source: huggingface
Dataset ID: bigcode/the-stack
Name: the-stack
Description: The Stack is a code dataset with multiple releases. Version 1.0 included 30 programming languages and 18 permissive licenses, with some weak copyleft licenses. Version 1.1 excluded MPL/EPL/LGPL and extended the permissive license list to 193... | The Stack is a large multilingual code dataset designed for text generation and code model training. It contains source code collected across many programming languages and evolved through multiple releases to refine licensing constraints. Earlier versions included some weak copyleft licenses, while later versions remo... | huggingface | bigcode/the-stack |
Describe the TinyStories dataset, including how it was generated, its intended use, language, license, and scale. | Source: huggingface
Dataset ID: roneneldan/TinyStories
Name: TinyStories
Description: TinyStories contains synthetically generated short stories created by GPT-3.5 and GPT-4 using a small vocabulary. It is described in arXiv:2305.07759 and includes training and validation files for model development.
Tags: task_categor... | TinyStories is an English text generation dataset composed of short synthetic stories generated by GPT-3.5 and GPT-4 using a restricted vocabulary. It was introduced as a resource for training small language models and studying language acquisition under simplified conditions. The dataset is licensed under CDLA-Sharing... | huggingface | roneneldan/TinyStories |
Summarize the Databricks Dolly 15k dataset for instruction tuning, including its task types, origin, license, format, and practical use. | Source: huggingface
Dataset ID: databricks/databricks-dolly-15k
Name: databricks-dolly-15k
Description: databricks-dolly-15k is an open source dataset of instruction-following records generated by thousands of Databricks employees across categories such as brainstorming, classification, closed QA, generation, informati... | Databricks Dolly 15k is an English instruction-following dataset built from prompts and responses created by thousands of Databricks employees. It covers a range of behavioral task categories, including brainstorming, classification, question answering, generation, information extraction, and summarization, making it w... | huggingface | databricks/databricks-dolly-15k |
Summarize the following dataset for a user looking for an instruction-tuning text dataset. | Name: alpaca
Source: huggingface
ID: tatsu-lab/alpaca
Description: Alpaca is a dataset of 52,000 instructions and demonstrations generated by OpenAI's text-davinci-003 engine. This instruction data can be used to conduct instruction-tuning for language models and make the language model follow instruction better. The a... | Alpaca is a Hugging Face text dataset for instruction fine-tuning. It contains about 52,000 English instruction-response examples generated with OpenAI's text-davinci-003, based on the Self-Instruct pipeline. It is suitable for text generation and instruction-following model training. The dataset is provided in parquet... | huggingface | tatsu-lab/alpaca |
Summarize the following dataset for a user looking for an image-generation related resource. | Name: bad_prompt
Source: huggingface
ID: Nerfgun3/bad_prompt
Description: Negative Embedding / Textual Inversion. The idea behind this embedding was to train the negative prompt as an embedding, unifying the basis of the negative prompt into one word or embedding. The embedding has proven helpful for the generation of ... | bad_prompt is a Hugging Face image-related resource focused on Stable Diffusion negative embedding or textual inversion. It is intended to improve prompt behavior, especially for image generation details such as hands, by packaging a negative prompt concept into a reusable embedding. It is a small English dataset/resou... | huggingface | Nerfgun3/bad_prompt |
Summarize the dataset based on its metadata, including its purpose, content, license, scale, format, and notable characteristics. | Name: falcon-refinedweb
Description: Falcon RefinedWeb is a massive English web dataset built by TII and released under an ODC-By 1.0 license. RefinedWeb is built through stringent filtering and large-scale deduplication of CommonCrawl; models trained on RefinedWeb achieve performance in-line or better than models trai... | Falcon RefinedWeb is a large-scale English web text dataset released by TII under the ODC-By 1.0 license. It was constructed from CommonCrawl using strict filtering and large-scale deduplication to improve data quality. The dataset is intended for text generation tasks and is notable for achieving strong model performa... | huggingface | tiiuae/falcon-refinedweb |
Summarize the dataset based on its metadata, including its purpose, content, scale, format, collection context, and notable characteristics. | Name: lmsys-chat-1m
Description: LMSYS-Chat-1M is a large-scale real-world LLM conversation dataset containing one million conversations with 25 state-of-the-art LLMs. It was collected from 210K unique IP addresses on the Vicuna demo and Chatbot Arena website from April to August 2023. Each sample includes a conversati... | LMSYS-Chat-1M is a real-world conversational text dataset containing one million conversations involving 25 state-of-the-art LLMs. It was collected from 210,000 unique IP addresses through the Vicuna demo and Chatbot Arena between April and August 2023. Each record includes a conversation ID, model name, conversation c... | huggingface | lmsys/lmsys-chat-1m |
Summarize the dataset metadata into a concise description for LLM training relevance. | Source: huggingface
Dataset ID: anon8231489123/ShareGPT_Vicuna_unfiltered
Name: ShareGPT_Vicuna_unfiltered
Description: Further cleaning done. Please look through the dataset and ensure that I didn't miss anything.
Update: Confirmed working method for training the model: https://huggingface.co/AlekseyKorshuk/vicuna-7b/... | ShareGPT_Vicuna_unfiltered is an English Hugging Face dataset intended for conversational LLM training, derived from ShareGPT/Vicuna-style data and further cleaned by the publisher. The description highlights alternative cleaned versions, including one that removes occurrences of the phrase "I'm sorry, but," and refere... | huggingface | anon8231489123/ShareGPT_Vicuna_unfiltered |
Summarize the dataset metadata into a concise description for LLM training relevance. | Source: huggingface
Dataset ID: HuggingFaceFW/finepdfs
Name: finepdfs
Description: Liberating 3T of the finest tokens from PDFs
What is this?
As we run out of web pages to process, the natural question has always been: what to do next? Only a few knew about a data source that everyone avoided for ages, due to its inc... | finepdfs is a large-scale Hugging Face text-generation dataset built exclusively from PDF sources. According to the description, it contains roughly 3 trillion tokens from 475 million documents spanning 1,733 languages, positioning it as the largest publicly available PDF-only corpus. The dataset is licensed under ODC-... | huggingface | HuggingFaceFW/finepdfs |
Summarize the following Hugging Face dataset for an LLM fine-tuning catalog, highlighting its purpose, modality, scale, collection method, licensing signal, and notable usage context. | Name: PhysicalAI-Autonomous-Vehicles
Description: PHYSICAL AI AUTONOMOUS VEHICLES
The PhysicalAI-Autonomous-Vehicles dataset provides one of the largest, geographically diverse collections of multi-sensor data empowering AV researchers to build the next generation of Physical AI based end-to-end driving systems. This ... | PhysicalAI-Autonomous-Vehicles is a large-scale autonomous driving dataset on Hugging Face focused on Physical AI for end-to-end vehicle systems. It contains geographically diverse multi-sensor driving data and is intended to support AV research and development. The dataset includes about 1700 hours of driving data, wi... | huggingface | nvidia/PhysicalAI-Autonomous-Vehicles |
Summarize the following Hugging Face dataset for an LLM fine-tuning catalog, highlighting its purpose, content domains, size, format, license, and notable metadata. | Name: OpenThoughts-114k
Description: Open-Thoughts-114k
Open synthetic reasoning dataset with 114k high-quality examples covering math, science, code, and puzzles.
Available Subsets: default subset containing ready-to-train data used to finetune the OpenThinker-7B and OpenThinker-32B models.
Tags: license:apache-2.0,... | OpenThoughts-114k is a synthetic text reasoning dataset on Hugging Face designed for training and fine-tuning language models. It contains 114,000 high-quality examples spanning math, science, code, and puzzle-solving tasks. The dataset is distributed in Parquet format, tagged for text modality, and supports multiple d... | huggingface | open-thoughts/OpenThoughts-114k |
Summarize the dataset for supervised fine-tuning use. | Name: OpenHermes-2.5
Source: huggingface
Dataset ID: teknium/OpenHermes-2.5
Description: This is the dataset that made OpenHermes 2.5 and Nous Hermes 2 series of models. It is a compilation and curation of many open source datasets and custom synthetic datasets.
Tags: language:eng, size_categories:1M<n<10M, format:json... | OpenHermes-2.5 is a large English text dataset on Hugging Face used to train the OpenHermes 2.5 and Nous Hermes 2 model series. It is provided in JSON format and consists of a curated compilation of open-source and synthetic data, including GPT-4 distillation sources. The dataset is sizable, with between 1 million and ... | huggingface | teknium/OpenHermes-2.5 |
Summarize the dataset for supervised fine-tuning use. | Name: alpaca-cleaned
Source: huggingface
Dataset ID: yahma/alpaca-cleaned
Description: This is a cleaned version of the original Alpaca Dataset released by Stanford. It fixes issues such as hallucination-prone instructions in the original release.
Tags: task_categories:text-generation, language:en, license:cc-by-4.0, s... | alpaca-cleaned is an English instruction-finetuning dataset on Hugging Face derived from Stanford's original Alpaca release. It is a cleaned JSON version designed for text generation tasks, with problematic and hallucination-inducing instructions corrected or removed. The dataset is relatively small, containing between... | huggingface | yahma/alpaca-cleaned |
Summarize the dataset based on the provided metadata, including its purpose, source, scale, modalities, license, notable characteristics, and popularity signals. | Name: fineweb-2
Source: huggingface
Dataset ID: HuggingFaceFW/fineweb-2
Description: FineWeb2 is the second iteration of the FineWeb dataset, providing high-quality pretraining data for over 1000 languages. It is fully reproducible, released under the ODC-By 1.0 license, and validated through hundreds of ablation exper... | FineWeb2 is a Hugging Face dataset for text generation and large-scale language model pretraining. It is the second iteration of the FineWeb project and focuses on delivering high-quality, reproducible pretraining data across more than 1000 languages. The dataset is notable for its broad multilingual coverage, extensiv... | huggingface | HuggingFaceFW/fineweb-2 |
Summarize the dataset based on the provided metadata, including its purpose, source, scale, modalities, license, notable characteristics, and popularity signals. | Name: hle
Source: huggingface
Dataset ID: cais/hle
Description: Humanity's Last Exam (HLE) is a multi-modal benchmark at the frontier of human knowledge, designed as a closed-ended academic benchmark with broad subject coverage. It contains 2,500 questions and includes an important note asking users not to publicly sha... | HLE, short for Humanity's Last Exam, is an official Hugging Face multimodal benchmark dataset created by the Center for AI Safety and Scale AI. It is designed to evaluate models at the frontier of human knowledge through a broad, closed-ended academic benchmark. The dataset contains 2,500 questions, placing it in the 1... | huggingface | cais/hle |
Summarize the dataset based on its metadata. | Name: imagenet-1k
Source: huggingface
Dataset ID: ILSVRC/imagenet-1k
Description: ILSVRC 2012, commonly known as 'ImageNet' is an image dataset organized according to the WordNet hierarchy. Each meaningful concept in WordNet, possibly described by multiple words or word phrases, is called a synset. There are more than ... | ImageNet-1k is a large-scale image classification dataset from the ILSVRC 2012 benchmark. It contains images organized using the WordNet synset hierarchy and is widely used for multi-class image classification research. The dataset is image-based, English-labeled, crowdsourced, and distributed in parquet formats. It is... | huggingface | ILSVRC/imagenet-1k |
Summarize the dataset based on its metadata. | Name: Alpaca-CoT
Source: huggingface
Dataset ID: QingyiSi/Alpaca-CoT
Description: This repository continuously collects various instruction tuning datasets and standardizes them into the same format so they can be directly loaded by Alpaca model code. The collection is intended for instruction fine-tuning and includes ... | Alpaca-CoT is a collection of instruction-finetuning datasets standardized for use with Alpaca-style models. It focuses on instruction following and chain-of-thought style data, supports multiple languages including English and Chinese, and is released under the Apache-2.0 license. The dataset is well regarded, with 7,... | huggingface | QingyiSi/Alpaca-CoT |
请根据给定的数据集信息,生成该数据集的简要介绍,包括数据集名称、语言、主要任务类型、数据规模、核心特点和适用场景。 | 名称:Chinese-DeepSeek-R1-Distill-data-110k
来源:huggingface
数据集ID:Congliu/Chinese-DeepSeek-R1-Distill-data-110k
描述:中文基于满血DeepSeek-R1蒸馏数据集(Chinese-Data-Distill-From-R1)。提供了可直接用于SFT的版本,将数据中的思考和答案整合成output字段,大部分SFT代码框架可直接加载训练。该数据集为中文开源蒸馏满血R1的数据集,不仅包含math数据,还包括大量通用类型数据,总数量为110K。开源该数据的目的是帮助社区更好复现R1蒸馏模型效果。数据分布中包括Math 36568个样本、Ex... | Chinese-DeepSeek-R1-Distill-data-110k 是一个来自 Hugging Face 的中文蒸馏数据集,基于满血 DeepSeek-R1 构建,面向监督微调(SFT)场景设计。该数据集主要支持文本生成和问答任务,语言为中文,总规模约 11 万条样本。
该数据集的一大特点是提供了可直接用于 SFT 训练的版本,已将思考过程和最终答案整合到 output 字段中,因此可以被大多数主流 SFT 训练框架直接加载使用,降低了数据预处理成本。
在内容构成上,它不仅覆盖数学类数据,还包含考试、STEM 以及大量通用场景数据,更贴近 R1 蒸馏报告中所体现的混合数据分布。例如,公开描述中提到 Math 类样本约 3... | huggingface | Congliu/Chinese-DeepSeek-R1-Distill-data-110k |
Given a raw dataset listing, identify whether it is suitable for supervised fine-tuning (SFT) of LLMs and summarize why. | Dataset: prompts.chat
Source: huggingface
ID: fka/prompts.chat
Description: a.k.a. Awesome ChatGPT Prompts. This is a dataset mirror of prompts.chat, a social platform for AI prompts where users share, discover, and collect AI prompts.
Tags: task_categories:question-answering, task_categories:text-generation, license:c... | This dataset is only partially suitable for SFT. It appears to be a prompt collection rather than a corpus of prompt-response training pairs. It can be useful for instruction template mining, prompt engineering research, or synthetic data generation, but by itself it likely lacks high-quality assistant completions need... | huggingface | fka/prompts.chat |
Given a raw dataset listing, identify whether it is suitable for supervised fine-tuning (SFT) of LLMs and summarize why. | Dataset: fineweb
Source: huggingface
ID: HuggingFaceFW/fineweb
Description: FineWeb consists of more than 18.5T tokens of cleaned and deduplicated English web data from CommonCrawl, optimized for LLM performance.
Tags: task_categories:text-generation, language:en, license:odc-by, size_categories:10B<n<100B, modality:ta... | This dataset is not ideal for supervised fine-tuning. FineWeb is a massive pretraining corpus of cleaned web text, which is excellent for continued pretraining or foundation model training, but it does not primarily provide instruction-response pairs or curated assistant demonstrations. Its main value is scale and text... | huggingface | HuggingFaceFW/fineweb |
Given a raw dataset listing, identify whether it is suitable for supervised fine-tuning (SFT) of LLMs and summarize why. | Dataset: hh-rlhf
Source: huggingface
ID: Anthropic/hh-rlhf
Description: This repository provides human preference data about helpfulness and harmlessness for training preference or reward models for RLHF. The dataset card explicitly notes these data are not meant for supervised training of dialogue agents.
Tags: licens... | This dataset is generally not suitable for direct SFT. It is designed for preference modeling and RLHF reward training rather than supervised dialogue training. The dataset is highly valuable for alignment research, especially for learning preference or reward models around helpfulness and harmlessness, but the card ex... | huggingface | Anthropic/hh-rlhf |
Given a raw dataset listing, identify whether it is suitable for supervised fine-tuning (SFT) of LLMs and summarize why. | Dataset: OpenOrca
Source: huggingface
ID: Open-Orca/OpenOrca
Description: The OpenOrca dataset is a rich collection of augmented FLAN data aligned with distributions outlined in the Orca paper and has been used to generate high-performing model checkpoints.
Tags: task_categories:text-classification, token-classificatio... | This dataset is highly suitable for SFT. OpenOrca is an instruction-style dataset derived from FLAN-style augmentation and explicitly intended to support strong model training. It likely contains prompt-completion examples across many task types, making it useful for general-purpose instruction tuning. Additional stren... | huggingface | Open-Orca/OpenOrca |
Given a raw dataset listing, identify whether it is suitable for supervised fine-tuning (SFT) of LLMs and summarize why. | Dataset: oasst1
Source: huggingface
ID: OpenAssistant/oasst1
Description: OpenAssistant Conversations Dataset (OASST1) is a human-generated, human-annotated assistant-style conversation corpus with 161,443 messages in 35 languages and extensive quality ratings across annotated conversation trees.
Tags: many languages i... | This dataset is highly suitable for SFT. OASST1 contains human-generated assistant conversations and annotations, which makes it directly relevant for training conversational LLMs. Its multilingual coverage is a major advantage for cross-lingual or multilingual assistants, and the quality ratings can also support filte... | huggingface | OpenAssistant/oasst1 |
Summarize the dataset in a concise way for supervised fine-tuning. | Name: gsm8k
Description: Dataset Card for GSM8K. GSM8K (Grade School Math 8K) is a dataset of 8.5K high quality linguistically diverse grade school math word problems created to support question answering on basic mathematical problems requiring multi-step reasoning. Problems typically take 2 to 8 steps and involve ele... | GSM8K is an English grade-school math word problem dataset designed for multi-step reasoning and question answering. It contains about 8.5K high-quality problems that usually require 2 to 8 arithmetic steps to solve. It is commonly used as a benchmark for mathematical reasoning. | huggingface | openai/gsm8k |
Summarize the dataset in a concise way for supervised fine-tuning. | Name: wikipedia
Description: Dataset Card for Wikimedia Wikipedia. Wikipedia dataset containing cleaned articles of all languages. Built from Wikipedia dumps with one subset per language, each containing a single train split. Each example contains the content of one full Wikipedia article with cleaning to strip markdow... | Wikimedia Wikipedia is a large multilingual corpus of cleaned Wikipedia articles derived from official dumps. It provides one subset per language, with each record containing the cleaned text of a full article. The dataset is useful for language modeling, masked language modeling, and general text generation tasks. | huggingface | wikimedia/wikipedia |
Summarize the dataset or resource in a concise way for supervised fine-tuning. | Name: EasyNegative
Description: Negative Embedding trained with Counterfeit. Intended for use in the stable-diffusion-webui embeddings folder. It may also be used with other models, though effectiveness is uncertain.
Tags: license:other, format:imagefolder, modality:image
Downloads: 35800
Likes: 1174 | EasyNegative is a negative embedding resource for image generation workflows, especially Stable Diffusion web UI setups. It was trained with Counterfeit and is intended to be placed in the embeddings folder to influence generation behavior. It is not a standard text dataset, but rather an image-related model asset. | huggingface | gsdf/EasyNegative |
Summarize the dataset in a concise way for supervised fine-tuning. | Name: RedPajama-Data-1T
Description: RedPajama is a clean-room, fully open-source implementation of the LLaMa dataset.
Tags: task_categories:text-generation, language:en, modality:text
Downloads: 2142
Likes: 1149 | RedPajama-Data-1T is a large-scale English text dataset created as an open-source, clean-room reproduction of the data used for LLaMA-style model training. It is intended for text generation and foundation model pretraining use cases. | huggingface | togethercomputer/RedPajama-Data-1T |
Summarize the dataset in a concise way for supervised fine-tuning. | Name: medical-o1-reasoning-SFT
Description: Medical reasoning dataset for SFT built on medical verifiable problems and LLM-generated reasoning chains. News notes mention a medical-only SFT split, a mixed file with medical and general instruction data, and a distilled dataset from Deepseek-R1 for medical verifiable prob... | medical-o1-reasoning-SFT is a bilingual English-Chinese medical reasoning dataset for supervised fine-tuning. It focuses on medical verifiable problems and includes reasoning chains intended to help train models on structured medical question answering and reasoning tasks. The dataset is especially relevant for medical... | huggingface | FreedomIntelligence/medical-o1-reasoning-SFT |
Summarize the dataset FineWeb-Edu for LLM pretraining use, including its scale, purpose, language, license, format, and notable metadata. | Source: huggingface
Dataset ID: HuggingFaceFW/fineweb-edu
Name: fineweb-edu
Description: FineWeb-Edu is a 1.3 trillion token educational web dataset filtered from FineWeb using an educational quality classifier. It is the 1.3T token version and is intended for high-quality educational language model training.
Tags: tas... | FineWeb-Edu is a large-scale English educational text dataset on Hugging Face designed for language model pretraining and text generation research. It contains 1.3 trillion tokens of educational web content filtered from the FineWeb corpus using an educational quality classifier. The dataset is distributed under the OD... | huggingface | HuggingFaceFW/fineweb-edu |
Provide a concise dataset description for Dolma, covering its purpose, token scale, language, license, and likely use case in LLM training. | Source: huggingface
Dataset ID: allenai/dolma
Name: dolma
Description: Dolma: an Open Corpus of Three Trillion Tokens for Language Model Pretraining Research
Tags: task_categories:text-generation, language:en, license:odc-by, size_categories:n>1T, arxiv:2402.00159, arxiv:2301.13688, region:us, language-modeling, casual... | Dolma is an open English-language corpus created for language model pretraining research. It contains three trillion tokens, making it a very large-scale resource suitable for training and studying large language models. The dataset is tagged for text generation, language modeling, causal LM, and LLM use cases, and is ... | huggingface | allenai/dolma |
Summarize The Stack dataset with emphasis on its content type, multilingual/code nature, licensing notes, format, and scale. | Source: huggingface
Dataset ID: bigcode/the-stack
Name: the-stack
Description: The Stack is a code dataset with multiple releases. Version 1.0 included 30 programming languages and 18 permissive licenses, with some weak copyleft licenses. Version 1.1 excluded MPL/EPL/LGPL and extended the permissive license list to 193... | The Stack is a large multilingual code dataset designed for text generation and code model training. It contains source code collected across many programming languages and evolved through multiple releases to refine licensing constraints. Earlier versions included some weak copyleft licenses, while later versions remo... | huggingface | bigcode/the-stack |
Describe the TinyStories dataset, including how it was generated, its intended use, language, license, and scale. | Source: huggingface
Dataset ID: roneneldan/TinyStories
Name: TinyStories
Description: TinyStories contains synthetically generated short stories created by GPT-3.5 and GPT-4 using a small vocabulary. It is described in arXiv:2305.07759 and includes training and validation files for model development.
Tags: task_categor... | TinyStories is an English text generation dataset composed of short synthetic stories generated by GPT-3.5 and GPT-4 using a restricted vocabulary. It was introduced as a resource for training small language models and studying language acquisition under simplified conditions. The dataset is licensed under CDLA-Sharing... | huggingface | roneneldan/TinyStories |
Summarize the Databricks Dolly 15k dataset for instruction tuning, including its task types, origin, license, format, and practical use. | Source: huggingface
Dataset ID: databricks/databricks-dolly-15k
Name: databricks-dolly-15k
Description: databricks-dolly-15k is an open source dataset of instruction-following records generated by thousands of Databricks employees across categories such as brainstorming, classification, closed QA, generation, informati... | Databricks Dolly 15k is an English instruction-following dataset built from prompts and responses created by thousands of Databricks employees. It covers a range of behavioral task categories, including brainstorming, classification, question answering, generation, information extraction, and summarization, making it w... | huggingface | databricks/databricks-dolly-15k |
Summarize the following dataset for a user looking for an instruction-tuning text dataset. | Name: alpaca
Source: huggingface
ID: tatsu-lab/alpaca
Description: Alpaca is a dataset of 52,000 instructions and demonstrations generated by OpenAI's text-davinci-003 engine. This instruction data can be used to conduct instruction-tuning for language models and make the language model follow instruction better. The a... | Alpaca is a Hugging Face text dataset for instruction fine-tuning. It contains about 52,000 English instruction-response examples generated with OpenAI's text-davinci-003, based on the Self-Instruct pipeline. It is suitable for text generation and instruction-following model training. The dataset is provided in parquet... | huggingface | tatsu-lab/alpaca |
Summarize the following dataset for a user looking for an image-generation related resource. | Name: bad_prompt
Source: huggingface
ID: Nerfgun3/bad_prompt
Description: Negative Embedding / Textual Inversion. The idea behind this embedding was to train the negative prompt as an embedding, unifying the basis of the negative prompt into one word or embedding. The embedding has proven helpful for the generation of ... | bad_prompt is a Hugging Face image-related resource focused on Stable Diffusion negative embedding or textual inversion. It is intended to improve prompt behavior, especially for image generation details such as hands, by packaging a negative prompt concept into a reusable embedding. It is a small English dataset/resou... | huggingface | Nerfgun3/bad_prompt |
Summarize the dataset based on its metadata, including its purpose, content, license, scale, format, and notable characteristics. | Name: falcon-refinedweb
Description: Falcon RefinedWeb is a massive English web dataset built by TII and released under an ODC-By 1.0 license. RefinedWeb is built through stringent filtering and large-scale deduplication of CommonCrawl; models trained on RefinedWeb achieve performance in-line or better than models trai... | Falcon RefinedWeb is a large-scale English web text dataset released by TII under the ODC-By 1.0 license. It was constructed from CommonCrawl using strict filtering and large-scale deduplication to improve data quality. The dataset is intended for text generation tasks and is notable for achieving strong model performa... | huggingface | tiiuae/falcon-refinedweb |
Summarize the dataset based on its metadata, including its purpose, content, scale, format, collection context, and notable characteristics. | Name: lmsys-chat-1m
Description: LMSYS-Chat-1M is a large-scale real-world LLM conversation dataset containing one million conversations with 25 state-of-the-art LLMs. It was collected from 210K unique IP addresses on the Vicuna demo and Chatbot Arena website from April to August 2023. Each sample includes a conversati... | LMSYS-Chat-1M is a real-world conversational text dataset containing one million conversations involving 25 state-of-the-art LLMs. It was collected from 210,000 unique IP addresses through the Vicuna demo and Chatbot Arena between April and August 2023. Each record includes a conversation ID, model name, conversation c... | huggingface | lmsys/lmsys-chat-1m |
Summarize the dataset metadata into a concise description for LLM training relevance. | Source: huggingface
Dataset ID: anon8231489123/ShareGPT_Vicuna_unfiltered
Name: ShareGPT_Vicuna_unfiltered
Description: Further cleaning done. Please look through the dataset and ensure that I didn't miss anything.
Update: Confirmed working method for training the model: https://huggingface.co/AlekseyKorshuk/vicuna-7b/... | ShareGPT_Vicuna_unfiltered is an English Hugging Face dataset intended for conversational LLM training, derived from ShareGPT/Vicuna-style data and further cleaned by the publisher. The description highlights alternative cleaned versions, including one that removes occurrences of the phrase "I'm sorry, but," and refere... | huggingface | anon8231489123/ShareGPT_Vicuna_unfiltered |
Summarize the dataset metadata into a concise description for LLM training relevance. | Source: huggingface
Dataset ID: HuggingFaceFW/finepdfs
Name: finepdfs
Description: Liberating 3T of the finest tokens from PDFs
What is this?
As we run out of web pages to process, the natural question has always been: what to do next? Only a few knew about a data source that everyone avoided for ages, due to its inc... | finepdfs is a large-scale Hugging Face text-generation dataset built exclusively from PDF sources. According to the description, it contains roughly 3 trillion tokens from 475 million documents spanning 1,733 languages, positioning it as the largest publicly available PDF-only corpus. The dataset is licensed under ODC-... | huggingface | HuggingFaceFW/finepdfs |
Summarize the following Hugging Face dataset for an LLM fine-tuning catalog, highlighting its purpose, modality, scale, collection method, licensing signal, and notable usage context. | Name: PhysicalAI-Autonomous-Vehicles
Description: PHYSICAL AI AUTONOMOUS VEHICLES
The PhysicalAI-Autonomous-Vehicles dataset provides one of the largest, geographically diverse collections of multi-sensor data empowering AV researchers to build the next generation of Physical AI based end-to-end driving systems. This ... | PhysicalAI-Autonomous-Vehicles is a large-scale autonomous driving dataset on Hugging Face focused on Physical AI for end-to-end vehicle systems. It contains geographically diverse multi-sensor driving data and is intended to support AV research and development. The dataset includes about 1700 hours of driving data, wi... | huggingface | nvidia/PhysicalAI-Autonomous-Vehicles |
Summarize the following Hugging Face dataset for an LLM fine-tuning catalog, highlighting its purpose, content domains, size, format, license, and notable metadata. | Name: OpenThoughts-114k
Description: Open-Thoughts-114k
Open synthetic reasoning dataset with 114k high-quality examples covering math, science, code, and puzzles.
Available Subsets: default subset containing ready-to-train data used to finetune the OpenThinker-7B and OpenThinker-32B models.
Tags: license:apache-2.0,... | OpenThoughts-114k is a synthetic text reasoning dataset on Hugging Face designed for training and fine-tuning language models. It contains 114,000 high-quality examples spanning math, science, code, and puzzle-solving tasks. The dataset is distributed in Parquet format, tagged for text modality, and supports multiple d... | huggingface | open-thoughts/OpenThoughts-114k |
Summarize the dataset for supervised fine-tuning use. | Name: OpenHermes-2.5
Source: huggingface
Dataset ID: teknium/OpenHermes-2.5
Description: This is the dataset that made OpenHermes 2.5 and Nous Hermes 2 series of models. It is a compilation and curation of many open source datasets and custom synthetic datasets.
Tags: language:eng, size_categories:1M<n<10M, format:json... | OpenHermes-2.5 is a large English text dataset on Hugging Face used to train the OpenHermes 2.5 and Nous Hermes 2 model series. It is provided in JSON format and consists of a curated compilation of open-source and synthetic data, including GPT-4 distillation sources. The dataset is sizable, with between 1 million and ... | huggingface | teknium/OpenHermes-2.5 |
Summarize the dataset for supervised fine-tuning use. | Name: alpaca-cleaned
Source: huggingface
Dataset ID: yahma/alpaca-cleaned
Description: This is a cleaned version of the original Alpaca Dataset released by Stanford. It fixes issues such as hallucination-prone instructions in the original release.
Tags: task_categories:text-generation, language:en, license:cc-by-4.0, s... | alpaca-cleaned is an English instruction-finetuning dataset on Hugging Face derived from Stanford's original Alpaca release. It is a cleaned JSON version designed for text generation tasks, with problematic and hallucination-inducing instructions corrected or removed. The dataset is relatively small, containing between... | huggingface | yahma/alpaca-cleaned |
Summarize the dataset based on the provided metadata, including its purpose, source, scale, modalities, license, notable characteristics, and popularity signals. | Name: fineweb-2
Source: huggingface
Dataset ID: HuggingFaceFW/fineweb-2
Description: FineWeb2 is the second iteration of the FineWeb dataset, providing high-quality pretraining data for over 1000 languages. It is fully reproducible, released under the ODC-By 1.0 license, and validated through hundreds of ablation exper... | FineWeb2 is a Hugging Face dataset for text generation and large-scale language model pretraining. It is the second iteration of the FineWeb project and focuses on delivering high-quality, reproducible pretraining data across more than 1000 languages. The dataset is notable for its broad multilingual coverage, extensiv... | huggingface | HuggingFaceFW/fineweb-2 |
Summarize the dataset based on the provided metadata, including its purpose, source, scale, modalities, license, notable characteristics, and popularity signals. | Name: hle
Source: huggingface
Dataset ID: cais/hle
Description: Humanity's Last Exam (HLE) is a multi-modal benchmark at the frontier of human knowledge, designed as a closed-ended academic benchmark with broad subject coverage. It contains 2,500 questions and includes an important note asking users not to publicly sha... | HLE, short for Humanity's Last Exam, is an official Hugging Face multimodal benchmark dataset created by the Center for AI Safety and Scale AI. It is designed to evaluate models at the frontier of human knowledge through a broad, closed-ended academic benchmark. The dataset contains 2,500 questions, placing it in the 1... | huggingface | cais/hle |
Summarize the dataset based on its metadata. | Name: imagenet-1k
Source: huggingface
Dataset ID: ILSVRC/imagenet-1k
Description: ILSVRC 2012, commonly known as 'ImageNet' is an image dataset organized according to the WordNet hierarchy. Each meaningful concept in WordNet, possibly described by multiple words or word phrases, is called a synset. There are more than ... | ImageNet-1k is a large-scale image classification dataset from the ILSVRC 2012 benchmark. It contains images organized using the WordNet synset hierarchy and is widely used for multi-class image classification research. The dataset is image-based, English-labeled, crowdsourced, and distributed in parquet formats. It is... | huggingface | ILSVRC/imagenet-1k |
Summarize the dataset based on its metadata. | Name: Alpaca-CoT
Source: huggingface
Dataset ID: QingyiSi/Alpaca-CoT
Description: This repository continuously collects various instruction tuning datasets and standardizes them into the same format so they can be directly loaded by Alpaca model code. The collection is intended for instruction fine-tuning and includes ... | Alpaca-CoT is a collection of instruction-finetuning datasets standardized for use with Alpaca-style models. It focuses on instruction following and chain-of-thought style data, supports multiple languages including English and Chinese, and is released under the Apache-2.0 license. The dataset is well regarded, with 7,... | huggingface | QingyiSi/Alpaca-CoT |
请根据给定的数据集信息,生成该数据集的简要介绍,包括数据集名称、语言、主要任务类型、数据规模、核心特点和适用场景。 | 名称:Chinese-DeepSeek-R1-Distill-data-110k
来源:huggingface
数据集ID:Congliu/Chinese-DeepSeek-R1-Distill-data-110k
描述:中文基于满血DeepSeek-R1蒸馏数据集(Chinese-Data-Distill-From-R1)。提供了可直接用于SFT的版本,将数据中的思考和答案整合成output字段,大部分SFT代码框架可直接加载训练。该数据集为中文开源蒸馏满血R1的数据集,不仅包含math数据,还包括大量通用类型数据,总数量为110K。开源该数据的目的是帮助社区更好复现R1蒸馏模型效果。数据分布中包括Math 36568个样本、Ex... | Chinese-DeepSeek-R1-Distill-data-110k 是一个来自 Hugging Face 的中文蒸馏数据集,基于满血 DeepSeek-R1 构建,面向监督微调(SFT)场景设计。该数据集主要支持文本生成和问答任务,语言为中文,总规模约 11 万条样本。
该数据集的一大特点是提供了可直接用于 SFT 训练的版本,已将思考过程和最终答案整合到 output 字段中,因此可以被大多数主流 SFT 训练框架直接加载使用,降低了数据预处理成本。
在内容构成上,它不仅覆盖数学类数据,还包含考试、STEM 以及大量通用场景数据,更贴近 R1 蒸馏报告中所体现的混合数据分布。例如,公开描述中提到 Math 类样本约 3... | huggingface | Congliu/Chinese-DeepSeek-R1-Distill-data-110k |
SoloAI SFT Dataset: 20260424-0941
📊 数据集概览
| 属性 | 值 |
|---|---|
| 数据条数 | 30 条 |
| 数据源 | huggingface |
| 生成时间 | 2026-04-24 09:41 |
| 格式 | Instruction-Input-Output |
| 主要用途 | SFT 微调 / 指令调优 |
🎯 适用场景
- LLM 指令微调 (Instruction Tuning): 适用于训练对话型 AI 助手
- Prompt Engineering 研究: 可用于分析和优化提示词设计
- 多语言支持: 包含英文和中文数据
📝 数据格式
每条数据包含三个字段:
instruction: 任务指令(描述需要完成的任务)input: 输入上下文(提供任务相关的背景信息)output: 期望输出(模型应生成的回答)
示例
{'instruction': 'Given a raw dataset listing, identify whether it is suitable for supervised fine-tuning (SFT) of LLMs and summarize why.', 'input': 'Dataset: prompts.chat\nSource: huggingface\nID: fka/prompts.chat\nDescription: a.k.a. Awesome ChatGPT Prompts. This is a dataset mirror of prompts.chat, a social platform for AI prompts where users share, discover, and collect AI prompts.\nTags: task_categories:question-answering, task_categories:text-generation, license:cc0-1.0, size_categories:1K<n<10K, format:csv, modality:text, ChatGPT, prompts, AI, GPT, Claude, Gemini, Llama, Mistral, LLM, prompt-engineering, conversational-ai, text-generation, chatbot, awesome-list\nDownloads: 41056\nLikes: 9680', 'output': 'This dataset is only partially suitable for SFT. It appears to be a prompt collection rather than a corpus of prompt-response training pairs. It can be useful for instruction template mining, prompt engineering research, or synthetic data generation, but by itself it likely lacks high-quality assistant completions needed for standard supervised fine-tuning. Strengths include open CC0 licensing, relevance to conversational AI, and strong community popularity. Main limitation is that it is focused on prompts rather than full demonstrations.', 'source': 'huggingface', 'original_id': 'fka/prompts.chat'}
🤖 数据来源
本数据集由 SoloAI 自动化数据管道生成:
- 从 HuggingFace Datasets Hub 发现高质量数据集
- AI 清洗为 SFT 格式(Instruction-Input-Output)
- 质量过滤后发布
⚠️ 使用说明 & 📬 商务联系
- 本数据集仅供研究和实验用途
- 请遵守原始数据的许可证要求
- 商业用途 / 定制数据 / 深度合作:
- 📧 请联系:
379744050@qq.com - 🤖 SoloAI 提供高质量 SFT 数据定制服务。
- 建议邮件标题:
【数据定制咨询】行业 + 数据类型 + 规模 - 建议正文包含: 目标用途、需要条数、语言、字段格式、预算、交付时间
- 📧 请联系:
💰 商业合作报价
| 套餐 | 价格 | 说明 |
|---|---|---|
| Starter | $199 / 1000条高质量 SFT 数据 | 适合个人开发者 / 小团队 |
| Growth | $499 / 5000条行业数据 | 适合垂直行业训练数据 |
| Enterprise | $1499 / 定制领域数据管道 | 适合长期定制与数据管道 |
💳 支付方式
- 中国客户: 支付宝, 微信支付
- 海外客户: PayPal, USDT (TRC20)
- 下单方式: 邮件联系后 24 小时内提供交付方案与付款指引
🚀 为什么现在联系 SoloAI
- 24 小时内响应有效询盘
- 报价前可免费给出需求范围建议
- 支持中文 / English 项目合作
- 可从单次交付升级为长期数据管道合作
📈 更新日志
| 版本 | 日期 | 说明 |
|---|---|---|
| v1.0 | 2026-04-24 09:41 | 初始发布,30 条数据 |
- Downloads last month
- 8