datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
DeepMath-103K
DeepMath-103K
🔥 News
May 8, 2025: We found that 48 samples contained hints that revealed the answers. The relevant questions have now been revised to remove the leaked answers.
April 14, 2025: We release DeepMath-103K, a large-scale dataset featuring challenging, verifiable, and decontaminated math problems tailored for RL and SFT. We open source:… See the full description on the dataset page: https://huggingface.co/datasets/zwhe99/DeepMath-103K.DeepSeek-v4-Pro-AgentThis dataset was generated using teich by TeichAI
Prepare these datasets for supervised fine-tuning in just a few lines of code — see the Conversion section below.
DeepSeek v4 Pro Agent Traces
This directory contains raw agent trace files generated by teich.
All assistant responses were generated by deepseek/deepseek-v4-pro.
JSONL files: 4006
Training-ready tools
A complete configured tools schema snapshot is embedded in the collapsed section at the bottom of… See the full description on the dataset page: https://huggingface.co/datasets/TeichAI/DeepSeek-v4-Pro-Agent.pg19This repository contains the PG-19 language modeling benchmark.
It includes a set of books extracted from the Project Gutenberg books library, that were published before 1919.
It also contains metadata of book titles and publication dates.
PG-19 is over double the size of the Billion Word benchmark and contains documents that are 20X longer, on average, than the WikiText long-range language modelling benchmark.
Books are partitioned into a train, validation, and test set. Book metadata is stored in metadata.csv which contains (book_id, short_book_title, publication_date).
Unlike prior benchmarks, we do not constrain the vocabulary size --- i.e. mapping rare words to an UNK token --- but instead release the data as an open-vocabulary benchmark. The only processing of the text that has been applied is the removal of boilerplate license text, and the mapping of offensive discriminatory words as specified by Ofcom to placeholder tokens. Users are free to model the data at the character-level, subword-level, or via any mechanism that can model an arbitrary string of text.
To compare models we propose to continue measuring the word-level perplexity, by calculating the total likelihood of the dataset (via any chosen subword vocabulary or character-based scheme) divided by the number of tokens --- specified below in the dataset statistics table.
One could use this dataset for benchmarking long-range language models, or use it to pre-train for other natural language processing tasks which require long-range reasoning, such as LAMBADA or NarrativeQA. We would not recommend using this dataset to train a general-purpose language model, e.g. for applications to a production-system dialogue agent, due to the dated linguistic style of old texts and the inherent biases present in historical writing.MacroLens
MacroLens
A benchmarking corpus for contextual financial reasoning under macroeconomic scenarios across 4,416 U.S. small- and micro-cap equities (2021-01-04 — 2026-03-31). MacroLens unifies seven tasks over a single point-in-time panel: contextual time-series forecasting, public valuation, financial-statement generation, scenario-conditioned return forecasting, private-company valuation, generator evaluation from natural-language descriptions, and real-estate valuation.
Task… See the full description on the dataset page: https://huggingface.co/datasets/DeepAuto-AI/MacroLens.DeepReview-Bench
DeepReview-Bench
A benchmark package built from DeepReview-13K. Each completed paper directory
contains the selected review-time PDF, a Markdown conversion generated from the
DeepReview-13K embedded source text, human-review data, and provenance metadata.
Contents
papers/<paper_id>/:
file
description
paper.pdf
selected PDF revision for review-time use
paper.md
Markdown converted from DeepReview-13K embedded source text
paper.source.tex
embedded… See the full description on the dataset page: https://huggingface.co/datasets/cmwqfcmwqf/DeepReview-Bench.DeepPlanning
DeepPlanning: Benchmarking Long-Horizon Agentic Planning with Verifiable Constraints
DeepPlanningBench is a challenging benchmark for evaluating long-horizon agentic planning capabilities of large language models (LLMs) with verifiable constraints. It features realistic multi-day travel planning and multi-product shopping tasks that require proactive information acquisition, local constrained reasoning, and global constrained optimization.
🌐 Website:… See the full description on the dataset page: https://huggingface.co/datasets/Qwen/DeepPlanning.DeepSWEGym2
Dataset Description
This dataset is a filtered and deduplicated version of a merge containing many high quality SWE datasets, it aims to improve benchmark results on DeepSWE-style problems, benchmarks, and general coding skills.
It is specifically filtered for rows with complex/long code problems in the original datasets, having an average row size of 214.19kb, a total uncompressed size of 17.56GB, and a total of 85974 examples.
Dataset Details
Curated by:… See the full description on the dataset page: https://huggingface.co/datasets/MoreThought/DeepSWEGym2.AM-DeepSeek-R1-Distilled-1.4MFor more open-source datasets, models, and methodologies, please visit our GitHub repository.
AM-DeepSeek-R1-Distilled-1.4M is a large-scale general reasoning task dataset composed of
high-quality and challenging reasoning problems. These problems are collected from numerous
open-source datasets, semantically deduplicated, and cleaned to eliminate test set contamination.
All responses in the dataset are distilled from the reasoning model (mostly DeepSeek-R1) and have undergone
rigorous… See the full description on the dataset page: https://huggingface.co/datasets/a-m-team/AM-DeepSeek-R1-Distilled-1.4M.AM-DeepSeek-Distilled-40MFor more open-source datasets, models, and methodologies, please visit our GitHub repository and paper: DeepDistill: Enhancing LLM Reasoning Capabilities via Large-Scale Difficulty-Graded Data Training.
Due to certain constraints, we are only able to open-source a subset of the complete dataset.
Model Training Performance based on our complete dataset
On AIME 2024, our 72B model achieved a score of 79.2 using only supervised fine-tuning (SFT). The 32B model reached 75.8 and… See the full description on the dataset page: https://huggingface.co/datasets/a-m-team/AM-DeepSeek-Distilled-40M.deepseek-v2-codder-minecraft-apideepsql_training
SynSQL Data Processing
A Python tool for processing the SynSQL-2.5M dataset into optimized Parquet format for machine learning workflows.
The dataset is split into batches of 30K entries with chain of thought(COT) reasoning and the answer. This can then be preprocessed and used for training any reasoning model.
Dataset Acknowledgment
This project processes data from the SynSQL-2.5M dataset by seeklhy, which is licensed under Apache 2.0. We acknowledge and thank the… See the full description on the dataset page: https://huggingface.co/datasets/ameet/deepsql_training.SWE-Milestone-data
Software evolution itineraries (as Milestone DAGs) extracted from real-world repositories for AI agent evaluation. Used by SWE-Milestone. [Paper]
This dataset contains the metadata, task specifications (SRS documents), dependency graphs, and test classifications (e.g., fail_to_pass, pass_to_pass) needed to run SWE-Milestone evaluation trials.
Dataset Statistics
SWE-Milestone covers 7 real-world open-source repositories spanning 5 programming languages… See the full description on the dataset page: https://huggingface.co/datasets/DeepCommit-ai/SWE-Milestone-data.ArXivSignals-DeepSummaries
ArXivSignals DeepSummaries — Agent-Built Visual Paper Explainers
A continuously-updated, day-partitioned dataset of deep, visual summaries of
arXiv papers, each built by a coding agent working inside the paper's own
LaTeX source: the agent reads the full text, authors an editorial narrative as
a structured content spec, and the paper's real figures and tables
(extracted and rendered from the LaTeX, web-optimized) ride along as an
embedded, variable-length image array. The… See the full description on the dataset page: https://huggingface.co/datasets/taesiri/ArXivSignals-DeepSummaries.DeepSWEGym
Dataset Description
This dataset is a filtered version of all the SWE-bench/SWE-smith-lang datasets (expect php) merged together (originally 88k rows total), it aims to improve benchmark results on DeepSWE-style problems, benchmarks, and general coding skills.
It is specifically filtered for rows with complex/long code problems in the original datasets, having an average row size of 94.41kb, a total uncompressed size of 6.12GB, and a total of 64821 examples.
Dataset… See the full description on the dataset page: https://huggingface.co/datasets/MoreThought/DeepSWEGym.DeepSWEGym2-Ultra
Dataset Description
This dataset is an EXTREMELY filtered and deduplicated version of DeepSWE-Gym2-Edu, it aims to improve benchmark results on DeepSWE-style problems, benchmarks, and general coding skills.
It is specifically filtered for rows with complex/long code problems in the original datasets, having an average row size of 8.27MB, a total uncompressed size of 8.27GB, and a total of 1000 examples.
Dataset Details
Curated by: MoreThought
Funded by:… See the full description on the dataset page: https://huggingface.co/datasets/MoreThought/DeepSWEGym2-Ultra.DeepResearch-Bench-II-DatasetDeepSeek-v4-Pro-AgentThis dataset was generated using teich by TeichAI
Prepare these datasets for supervised fine-tuning in just a few lines of code — see the Conversion section below.
DeepSeek v4 Pro Agent Traces
This directory contains raw agent trace files generated by teich.
All assistant responses were generated by deepseek/deepseek-v4-pro.
JSONL files: 4006
Training-ready tools
A complete configured tools schema snapshot is embedded in the collapsed section at the bottom of… See the full description on the dataset page: https://huggingface.co/datasets/ronaldcmz/DeepSeek-v4-Pro-Agent.deepseek-v4-pro-0813-agentic
DeepSeek-V4-Pro 0813 Agentic (DS4)
A standalone, verifiable-first agentic training corpus: 19,072 training traces
plus 2,135 held-out evaluation rows (validation 1,070 / test 1,065), generated by
DeepSeek-V4-Pro 0813 (deepseek-v4-pro-0813, official API, thinking mode) across 13 verifiable task families,
each row admitted only after passing a deterministic programmatic verifier. The corpus is
designed to be directly usable for SFT, GRPO/RLVR, and NeMo Gym / NeMo RL
(verified… See the full description on the dataset page: https://huggingface.co/datasets/r0b0tlab/deepseek-v4-pro-0813-agentic.AM-DeepSeek-R1-0528-Distilled
📘 Dataset Summary
This dataset is a high-quality reasoning corpus distilled from DeepSeek-R1-0528, an improved version of the DeepSeek-R1 large language model. Compared to its initial release, DeepSeek-R1-0528 demonstrates significant advances in reasoning, instruction following, and multi-turn dialogue. Motivated by these improvements, we collected and distilled a diverse set of 2.6 million queries across multiple domains, using DeepSeek-R1-0528 as the teacher.
A notable… See the full description on the dataset page: https://huggingface.co/datasets/a-m-team/AM-DeepSeek-R1-0528-Distilled.DeepSeek-V4-Pro-Distilled-200K
DeepSeek‑V4‑Pro‑Distilled‑200K
High-quality Math & STEM reasoning distilled from DeepSeek‑V4‑Pro in Max mode
Reasoning traces · Proofs · Verification · Mathematics · Physics · Chemistry · Biology
Overview
DeepSeek‑V4‑Pro‑Distilled‑200K is a supervised fine-tuning collection of long-form mathematical and scientific reasoning. Its responses were generated with DeepSeek‑V4‑Pro in Max inference mode, then normalized into a compact conversational… See the full description on the dataset page: https://huggingface.co/datasets/Jackrong/DeepSeek-V4-Pro-Distilled-200K.DeepSWEGym2-Edu
Dataset Description
This dataset is a filtered and deduplicated version of a merge containing many high quality SWE datasets, it aims to improve benchmark results on DeepSWE-style problems, benchmarks, and general coding skills.
It is specifically filtered for rows with complex/long code problems in the original datasets, having an average row size of 291.74kb, a total uncompressed size of 14.93GB, and a total of 53649 examples.
Dataset Details
Curated by:… See the full description on the dataset page: https://huggingface.co/datasets/MoreThought/DeepSWEGym2-Edu.deepscaler-preview-verl
DeepScaleR-Preview VERL
📊 Dataset Summary
This dataset contains 35,789 mathematical reasoning problems in VERL format, processed from agentica-org/DeepScaleR-Preview-Dataset.
Key Features:
35,789 high-quality math problems
Converted to VERL format for reward modeling
Verified ground truth answers
Ready for reinforcement learning training
🔗 Source Dataset
Original Repository
Repository:… See the full description on the dataset page: https://huggingface.co/datasets/sungyub/deepscaler-preview-verl.Chinese-DeepSeek-R1-Distill-data-110k
中文基于满血DeepSeek-R1蒸馏数据集（Chinese-Data-Distill-From-R1）
🤗 Hugging Face | 🤖 ModelScope | 🚀 Github | 📑 Blog
注意：提供了直接SFT使用的版本，点击下载。将数据中的思考和答案整合成output字段，大部分SFT代码框架均可直接直接加载训练。
本数据集为中文开源蒸馏满血R1的数据集，数据集中不仅包含math数据，还包括大量的通用类型数据，总数量为110K。
为什么开源这个数据？
R1的效果十分强大，并且基于R1蒸馏数据SFT的小模型也展现出了强大的效果，但检索发现，大部分开源的R1蒸馏数据集均为英文数据集。 同时，R1的报告中展示，蒸馏模型中同时也使用了部分通用场景数据集。
为了帮助大家更好地复现R1蒸馏模型的效果，特此开源中文数据集。
该中文数据集中的数据分布如下：… See the full description on the dataset page: https://huggingface.co/datasets/Congliu/Chinese-DeepSeek-R1-Distill-data-110k.DeepSWEGym2-Full
Dataset Description
This dataset is a merge containing many high quality SWE datasets, it aims to improve benchmark results on DeepSWE-style problems, benchmarks, and general coding skills.
It is NOT specifically filtered for rows with complex/long code problems in the original datasets, despite still having an average row size of 169.08kb, a total uncompressed size of 19.80GB, and a total of 122791 examples.
Dataset Details
Curated by: MoreThought
Funded by:… See the full description on the dataset page: https://huggingface.co/datasets/MoreThought/DeepSWEGym2-Full.DeepSeek-V4-Distill-8000x
🐳 DeepSeek-V4-Distill-8100x
Dataset Summary
DeepSeek-V4-Distill-8100x is a supervised fine-tuning dataset for reasoning-oriented distillation. The question prompts come from Jackrong/GLM-5.1-Reasoning-1M-Cleaned, and the answers were generated by the teacher model DeepSeek-V4-Flash.
After the cleaning process, the released train split contains 7,716 high-quality JSONL examples.
[!NOTE]
The answer pool was cleaned to remove real-time questions… See the full description on the dataset page: https://huggingface.co/datasets/Jackrong/DeepSeek-V4-Distill-8000x.saas-sales-conversations
saas-sales-conversations
Dataset Description
This is a synthetic dataset of sales conversations for SaaS (Software as a Service) companies, designed for training sales conversion prediction models. The dataset was created following the methodology presented in "SalesRLAgent: A Reinforcement Learning Approach for Real-Time Sales Conversion Prediction and Optimization" (Nandakishor M, 2025).
The dataset contains realistic dialogues between sales representatives and… See the full description on the dataset page: https://huggingface.co/datasets/DeepMostInnovations/saas-sales-conversations.nemotron-deepseek-sft-mix-v2
Unified DeepSeek SFT Mix v2 (Nemotron-derived)
A training-ready SFT dataset that assembles the DeepSeek-generated portions of four public
NVIDIA Nemotron datasets and normalizes them into a single OpenAI-style message schema that
renders directly with the DeepSeek-V4 chat encoding.
This is the DeepSeek-V4-Pro generation companion to
ycchen/nemotron-deepseek-sft-mix
(v1, DeepSeek-V3.2 math / SWE / terminal). v2 adds DeepSeek-V4-Pro math & proofs,
scientific reasoning, and agentic… See the full description on the dataset page: https://huggingface.co/datasets/ycchen/nemotron-deepseek-sft-mix-v2.DeepSWEGym-Edu
Dataset Description
This dataset is a heavily filtered version of all the SWE-bench/SWE-smith-lang datasets (expect php) merged together (originally 88k rows total), it aims to improve benchmark results on DeepSWE-style problems, benchmarks, and general coding skills.
It is specifically filtered for rows with very complex/long code problems in the original datasets, having an average row size of 101.81kb, a total uncompressed size of 4.29GB, and 42529 examples total.… See the full description on the dataset page: https://huggingface.co/datasets/MoreThought/DeepSWEGym-Edu.cultura_ru_edu
Cultura-Ru-Edu
The Cultura-Ru-Edu dataset consists of Russian educational web pages filtered from the uonlp/CulturaX dataset.
The dataset creation was inspired by HuggingFaceFW/fineweb-edu, but with a focus on the Russian language.
By filtering the dataset based on educational criteria, the Cultura-Ru-Edu dataset is both high-quality and large enough to train a Russian-focused language model for tasks requiring knowledge of the world.
Dataset curation
To create this… See the full description on the dataset page: https://huggingface.co/datasets/deepvk/cultura_ru_edu.deepa2
deepa2 Datasets Collection
Dataset Summary
This is a growing, curated collection of deepa2 datasets, i.e. datasets that contain comprehensive logical analyses of argumentative texts. The collection comprises:
datasets that are built from existing NLP datasets by means of the deepa2 bake tool.
original deepa2 datasets specifically created for this collection.
The tool deepa2 serve may be used to render the data in this collection as text2text examples.… See the full description on the dataset page: https://huggingface.co/datasets/DebateLabKIT/deepa2.