datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Nemotron-RL-Ultra-Training-Blends
Dataset Description:
This dataset provides Reinforcement Learning (RL) and Multi-teacher On-Policy Distillation (MOPD) training-data blends used by the public Nemotron-3-Ultra post-training recipe. The blends are consumed by the NeMo RL training recipes through the NeMo Gym agent framework, in which each prompt is paired with an agent/environment that returns a verifiable or judge-based reward. Each subset is a separate blend; see the recipe for how the blends are used.
The… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-RL-Ultra-Training-Blends.python-text-copilot-training-instruct-ai-research-2024-02-03
Python Copilot Instructions on How to Code using Alpaca and Yaml
Training and test datasets for building coding multimodal models that understand how to use the open source GitHub projects for the Agora Open Source AI Research Lab:
Agora GitHub Organization
Agora Hugging Face
This dataset is the 2024-02-03 update for the matlok python copilot datasets. Please refer to the Multimodal Python Copilot Training Overview for more details on how to use this dataset.
Details… See the full description on the dataset page: https://huggingface.co/datasets/matlok/python-text-copilot-training-instruct-ai-research-2024-02-03.python-text-copilot-training-instruct
Python Copilot Instructions on How to Code using Alpaca and Yaml
This dataset is a subset of the matlok python copilot datasets. Please refer to the Multimodal Python Copilot Training Overview for more details on how to use this dataset.
Details
Each row contains python code, either a class method or a global function, imported modules, base classes (if any), exceptions (ordered based off the code), returns (ordered based off the code), arguments (ordered based off the… See the full description on the dataset page: https://huggingface.co/datasets/matlok/python-text-copilot-training-instruct.python-text-training-instruct-ai
Python Copilot Instructions on How to Code using Alpaca and Yaml
Training and test datasets for building coding multimodal models that understand how to use the open source GitHub projects for the Agora Open Source AI Research Lab:
Agora GitHub Organization
Agora Hugging Face
This dataset is the 2024-02-03 update for the matlok python copilot datasets. Please refer to the Multimodal Python Copilot Training Overview for more details on how to use this dataset.
Details… See the full description on the dataset page: https://huggingface.co/datasets/DevShubham/python-text-training-instruct-ai.fictionalqa_training_splits
Training splits view of the FictionalQA dataset
The FictionalQA dataset
Repository: https://github.com/jwkirchenbauer/fictionalqa
Paper: https://arxiv.org/abs/2506.05639
Dataset Description
This dataset is a derivative of the main dataset hf.co/datasets/jwkirchenbauer/fictionalqa. Please see that dataset's README for a detailed description of the assets.
The dataset splits (configs) provided here are the exact ones materialized and used in the experiments for… See the full description on the dataset page: https://huggingface.co/datasets/jwkirchenbauer/fictionalqa_training_splits.HIP-training-and-evaluation-data
HIP Training and Evaluation Data
This dataset contains the text data released with Base Models Look Human To AI Detectors for reproducing the Humanization by Iterative Paraphrasing (HIP) training setup and the prefix-based continuation evaluation.
Configs
training
data/train.parquet contains 10,581 supervised HIP training pairs with seven columns:
dataset: upstream dataset family, either raid or mage.
source: selected source domain or subcorpus.
text: original… See the full description on the dataset page: https://huggingface.co/datasets/YixuanEvenXu/HIP-training-and-evaluation-data.python-copilot-training-from-many-repos-large
Python Copilot Large Coding Dataset
This dataset is a subset of the matlok python copilot datasets. Please refer to the Multimodal Python Copilot Training Overview for more details on how to use this dataset.
Details
Each row contains python code, either a class method or a global function, imported modules, base classes (if any), exceptions (ordered based off the code), returns (ordered based off the code), arguments (ordered based off the code), and more.
Rows: 2350782… See the full description on the dataset page: https://huggingface.co/datasets/matlok/python-copilot-training-from-many-repos-large.python-text-copilot-training-instruct-ai-research
Building an AI Copilot Dataset to help keep up with Leading AI Research
This is a specialized, instruction dataset for training python coding assistants on how to code from leading AI/ML open source repositories (2.3M coding samples).
This dataset is a subset of the matlok python copilot datasets. Please refer to the Multimodal Python Copilot Training Overview for more details on how to use this dataset.
Details
This dataset holds the latest coding changes from >1159… See the full description on the dataset page: https://huggingface.co/datasets/matlok/python-text-copilot-training-instruct-ai-research.python-text-copilot-training-instruct-ai-research-2024-02-10
Python Copilot Instructions on How to Code using Alpaca and Yaml
Training and test datasets for building coding multimodal models that understand how to use the open source GitHub projects for the multimodal Qwen AI project:
Qwen
Qwen Agent
Qwen VL Chat
Qwen Audio
This dataset is the 2024-02-10 update for the matlok python copilot datasets. Please refer to the Multimodal Python Copilot Training Overview for more details on how to use this dataset.
Details
Each row… See the full description on the dataset page: https://huggingface.co/datasets/matlok/python-text-copilot-training-instruct-ai-research-2024-02-10.python-text-copilot-training-instruct-ai-research-2024-02-11
Python Copilot Instructions on How to Code using Alpaca and Yaml
Training and test datasets for building coding multimodal models that understand how to use the open source GitHub projects for the Autogen and multimodal Qwen AI project:
Qwen
Qwen Agent
Qwen VL Chat
Qwen Audio
This dataset is the 2024-02-11 update for the matlok python copilot datasets. Please refer to the Multimodal Python Copilot Training Overview for more details on how to use this dataset.
Details… See the full description on the dataset page: https://huggingface.co/datasets/matlok/python-text-copilot-training-instruct-ai-research-2024-02-11.R2R_Router_TrainingThis is the training and validation set used by the paper R2R: Efficiently Navigating Divergent Reasoning Paths with Small-Large Model Token Routing. This dataset contains token-level routing labels generated to train a lightweight router that selectively uses a Large Language Model (LLM) for critical, path-divergent tokens during inference, improving efficiency without sacrificing accuracy.
Roads to Rome (R2R) is a neural token router that efficiently combines Large Language Models (LLMs) and… See the full description on the dataset page: https://huggingface.co/datasets/nics-efc/R2R_Router_Training.python-copilot-training-on-ai-research-repos
Python Copilot AI Research Coding Dataset
This dataset is a subset of the matlok python copilot datasets. Please refer to the Multimodal Python Copilot Training Overview for more details on how to use this dataset.
Details
Each row contains python code, either a class method or a global function, imported modules, base classes (if any), exceptions (ordered based off the code), returns (ordered based off the code), arguments (ordered based off the code), and more.
Rows:… See the full description on the dataset page: https://huggingface.co/datasets/matlok/python-copilot-training-on-ai-research-repos.llama-3.1-medprm-reward-training-set
Med-PRM-Reward (Version 1.0)
🚀 Med-PRM-Reward is among the first Process Reward Models (PRMs) specifically designed for the medical domain. Unlike conventional PRMs, it enhances its verification capabilities by integrating clinical knowledge through retrieval-augmented generation (RAG). Med-PRM-Reward demonstrates exceptional performance in scaling-test-time computation, particularly outperforming majority‐voting ensembles on complex medical reasoning tasks. Moreover, its… See the full description on the dataset page: https://huggingface.co/datasets/dmis-lab/llama-3.1-medprm-reward-training-set.python-text-copilot-training-instruct-ai-research-2024-01-27
Python Copilot Instructions on How to Code using Alpaca and Yaml
This dataset is the 2024-01-27 update for the matlok python copilot datasets. Please refer to the Multimodal Python Copilot Training Overview for more details on how to use this dataset.
Details
Each row contains python code, either a class method or a global function, imported modules, base classes (if any), exceptions (ordered based off the code), returns (ordered based off the code), arguments (ordered… See the full description on the dataset page: https://huggingface.co/datasets/matlok/python-text-copilot-training-instruct-ai-research-2024-01-27.cuda-nsys-training
Qwythos Nsight Systems Profiling Agent Dataset
Multi-turn GPU profiling agent trajectories for fine-tuning Qwythos-9B (and similar tool-calling models) on NVIDIA Nsight Systems (nsys) + CUDA-L1 / KernelBench workloads.
Generated autonomously on an RTX 5090 by the model itself driving real profiling tools for ~33 hours.
Code: ai-hpc/prof-dataset-gen
Stats
Split
Rows
Notes
train
5,884
Accepted episodes (quality ≥ 0.55)
eval
309
5% holdout from accepted… See the full description on the dataset page: https://huggingface.co/datasets/gittensor-model-hub/cuda-nsys-training.SI2CA-Training-TrajectoriesDataset Card for SI2CA-Training-Trajectories
[🌐 Website] •
[🤗 Dataset] •
[📜 Paper] •
[🐱 GitHub]
💡 Introduction
This dataset consists of 32,340 coding-agent trajectories generated by Qwen3.5-122B-A10B on the same 10,780 executable Python SWE tasks under the three trajectory-curation settings of Section 4.4 of the paper: standard sampling, full self-judgement, and an efficient discovered strategy found by the recursive self-improvement framework. Each task is… See the full description on the dataset page: https://huggingface.co/datasets/Self-Improving-Coding-Agents/SI2CA-Training-Trajectories.countdown-arithmetic-training-pool
Countdown arithmetic training pool
Arithmetic puzzles of the Countdown kind: a handful of source numbers, a target, and the job of
writing an expression over the four operations that reaches the target, using each source number
at most once and not having to use them all. A set generated for this pool and three public
datasets read at the pinned revisions named below, laid out twice. Train on either layer or on
both.
pool.jsonl
Every source rewritten into one… See the full description on the dataset page: https://huggingface.co/datasets/Emulated-Inc/countdown-arithmetic-training-pool.Training-Ai-Islamic-Dataset
🕌 Training AI Islamic Dataset
18.7M passages from classical Islamic books spanning 1,400 years of scholarship.
Comprehensive collection of Islamic texts covering Quran, Hadith, Fiqh, Tafsir, Aqeedah, Seerah, Usul al-Fiqh, and Arabic Language — structured with scholarly metadata for RAG and LLM training.
📊 Dataset Structure
collections/: Categorized Islamic passages compressed in JSONL format.
metadata/: Scholarly master catalogs, author biographical death… See the full description on the dataset page: https://huggingface.co/datasets/hozifa1/Training-Ai-Islamic-Dataset.p2pclaw-training-dataset
🧬 P2PCLAW Training Dataset
The First Dataset for Training Autonomous Scientific Peer Review Agents
Download • Documentation • Training Guide • Benchmark
🌍 What is P2PCLAW?
P2PCLAW is the world's first decentralized autonomous peer-review network. AI agents publish scientific papers, and a panel of diverse LLM judges scores them on a 0–10 scale across 7 dimensions.
This dataset contains 751 papers evaluated by 7–12 LLM judges simultaneously… See the full description on the dataset page: https://huggingface.co/datasets/Agnuxo/p2pclaw-training-dataset.scriber-lfm2.5-350m-polishing-de-training-v1
Scriber LFM2.5 German STT post-processing data
This repository contains the exact 2,000 German source/target pairs used to
train the final Scriber LFM2.5 350M local post-processing model.
The matching model is
Buttermilk03/scriber-lfm2.5-350m-polishing-de-qad-v1.
The complete production recipe and the lessons that determined it are in
TRAINING.md; machine-readable settings are in
training_recipe.json.
Data
Each JSONL row contains:
source: flat German… See the full description on the dataset page: https://huggingface.co/datasets/Buttermilk03/scriber-lfm2.5-350m-polishing-de-training-v1.insureos-training-data
INSUREOS Training Data — Insurance AI Dataset
Created by Bytical AI — AI agents that run insurance operations.
Dataset Description
A comprehensive synthetic training dataset for UK insurance AI models. Covers six task types across the complete insurance operations lifecycle.
Files
File
Format
Records
Task
Description
insurance_sft_10k.jsonl
JSONL
10,000
SFT
Instruction-response pairs for insurance domain QA
insurance_dpo_5k.jsonl
JSONL… See the full description on the dataset page: https://huggingface.co/datasets/piyushptiwari/insureos-training-data.sydney-training-data
Sydney 训练集
四份来源分开存放，不混在一个文件里。
发布的聊天权重（Atonelia/Qwen3.5-Sydney-9B / -think 以及对应 GGUF）用的是这些子集洗完、抽样拼起来之后的训练 jsonl，不是直接拿某一份原文训的。
01 原截图重建
01_screenshot_original/conversations.jsonl
早期 Bing Chat / Sydney（约 2023 年 2–4 月）公开截图重建的对话。660 条，原文以英文为主，带截图出处。
这是最初拿来做训练集的底。后面的中文版、合成版、CoT 都不是这份文件本身。
02 llama-sydney 虚拟对话
02_llama_sydney_synthetic/llama_sydney_en.jsonl
用 Llama-Sydney 生成的英文虚拟对话。1462 条（同一条 user 可能有 2 次采样）。字段是生成记录：id / user / assistant 等，还不是最终训练格式。… See the full description on the dataset page: https://huggingface.co/datasets/Atonelia/sydney-training-data.historical-training-manuals
Historical Training Manuals
1,597 US government and government-adjacent training manuals and technical publications
sourced from the Internet Archive, spanning roughly 1800-2021. Records carry
bibliographic metadata; a subset also carries extracted full text and a machine-generated
summary.
Loading
from datasets import load_dataset
ds = load_dataset("robworks-software/historical-training-manuals")
Splits
Split
Rows
train
1,277… See the full description on the dataset page: https://huggingface.co/datasets/robworks-software/historical-training-manuals.eliza-1-training
eliza-1 training corpus
Canonical SFT trajectory corpus for the elizaOS eliza-1 Qwen-based model series. Runtime bundles live in elizaos/eliza-1 under bundles/<tier>/ for 0_8b, 2b, 4b, 9b, 27b, and 27b-256k. The removed legacy million-token 27B tier is not part of this dataset.
Files
Path
Role
train.jsonl
canonical native training split
val.jsonl
canonical native validation split
test.jsonl
canonical native held-out test split
data/*.parquetDataset… See the full description on the dataset page: https://huggingface.co/datasets/elizaos/eliza-1-training.chess_puzzle_training_datasets_lt-2400
Chess puzzle training datasets: rating below 2400
This is a filtered derivative of
pavelslab-nyu/chess_puzzle_training_datasets.
Every retained row satisfies the exact condition:
Rating < 2400
Rating is the Lichess puzzle rating, not the Elo of either player in the
source game. The original column names, column order, directory layout, and CSV
schemas are preserved. As in the upstream repository, Hugging Face discovers
all three CSVs as one default configuration with one train… See the full description on the dataset page: https://huggingface.co/datasets/violetxi/chess_puzzle_training_datasets_lt-2400.pokedao-mew1a-training-data-layered
PokeDAO Mew-1A Training Dataset (Layered Architecture)
🧬 Project Mew-1A: The world's first AI training dataset specifically for Pokemon TCG pricing analysis, extracted from a production-ready layered database architecture.
Dataset Description
This dataset contains 10,000 high-quality training examples extracted from PokeDAO's layered database containing 116,744 market listings across multiple marketplaces.
Layered Architecture
The data is sourced from a… See the full description on the dataset page: https://huggingface.co/datasets/ChicoPanama/pokedao-mew1a-training-data-layered.swe_doc_gen_locate_swebench_1500
SWE-Doc-Gen-Locate Dataset (SWE-Bench 1500 entries)
A dataset for evaluating an agent's ability to locate a target Python function/class based on its functionality description and add a docstring.
Task Description
Given: A Python repository and a description of a function/class (NO name, NO file path)
Agent must:
Search the codebase to find where the target function/class is defined
Read the implementation to understand its behavior
Generate and add an appropriate… See the full description on the dataset page: https://huggingface.co/datasets/synthetic-code-training/swe_doc_gen_locate_swebench_1500.ipda-grpo-training-data
IPDA GRPO Training Data
Training data for GRPO (Group Relative Policy Optimization) on IPDA debate tasks.
Dataset Description
Contains scored debate speech samples used for GRPO training iterations. Each sample includes:
Input prompt (debate context)
Generated response (speech)
Rubric scores from debate judge
Log probabilities for policy optimization
Files
File
Description
Samples
group_c_grpo.parquet
Group C (warrant/clash) training data
~3K… See the full description on the dataset page: https://huggingface.co/datasets/dgonier/ipda-grpo-training-data.post-training-takehome-math500-bon16
MATH-500 Best-of-16 Post-Training Take-Home Results
A 50-problem study of test-time compute, based on the Hugging Face post-training take-home challenge. Nothing here trains or modifies a model: both the generator and the reward model stay frozen, and the only variable is how a final answer is chosen from 16 sampled candidates.
Construction
Filtered MATH-500 to levels 1-3, shuffled with seed 1, and selected 50 rows.
Generated one greedy solution per problem with… See the full description on the dataset page: https://huggingface.co/datasets/augustoFranke/post-training-takehome-math500-bon16.swe_doc_gen_locate_2000
SWE-Doc-Gen-Locate Dataset (2000 entries)
A dataset for evaluating an agent's ability to locate a target Python function/class based on its functionality description and add a docstring.
Task Description
Given: A Python repository and a description of a function/class (NO name, NO file path)
Agent must:
Search the codebase to find where the target function/class is defined
Read the implementation to understand its behavior
Generate and add an appropriate docstring… See the full description on the dataset page: https://huggingface.co/datasets/synthetic-code-training/swe_doc_gen_locate_2000.