datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
eagle
Eagle 🦅: Ethical Dataset Given from Real Interactions
Introduction
This repository contains the Eagle dataset, which is an ethical dataset of real interactions between humans and ChatGPT. This dataset is created to evaluate social bias, opinion bias, toxic language, and morality in Large Language Models (LLMs).
If you use the Eagle dataset in your research, please cite the following:
@inproceedings{Eagle:arxiv:2024,
title={Eagle: Ethical Dataset Given from Real… See the full description on the dataset page: https://huggingface.co/datasets/MasahiroKaneko/eagle.EagleX-WorldContinued
Dataset Card for EagleX v2 Dataset
This dataset was used to train RWKV Eagle 7B for continued pretrain of 1.1T tokens (approximately) (boosting it to 2.25T) with the final model being released as RWKV EagleX v2.
Dataset Details
Dataset Description
EagleX-WorldContinued is a pretraining dataset built from many of our datasets over at Recursal AI + a few others.
Curated by: M8than, KaraKaraWitch, Darok
Funded by [optional]: Recursal.ai
Shared by [optional]:… See the full description on the dataset page: https://huggingface.co/datasets/RWKV/EagleX-WorldContinued.EagleSFT
Dataset Card for 🦅 EagleSFT
Dataset Summary
This dataset contains 536,231 pairs of human questions and machine-generated responses intended for supervised fine-tuning (SFT) of large language models. The dataset includes both Russian and English content, with linked IDs allowing for cross-lingual analysis. It was created by processing an initial collection of 739,732 human questions posed to LLMs, predominantly in Russian (about 99%) with a small portion in English (about… See the full description on the dataset page: https://huggingface.co/datasets/nyuuzyou/EagleSFT.multireward-grpo-gsm8k-rewards-qwen2.5-7b
Multi-Reward GRPO — GSM8K Rewards (Qwen2.5-7B-Instruct)
Raw rollout-level reward observations from the empirical Section of
"Conditioned Multi-Reward Advantage Estimation: A Finite-Sample Analysis".
This is the data that produced the headline Theorem 3 (correlation-dependent
MSE floor) and Proposition 4 (sign-changing conditioning bias) figures on
real LLM rollouts. Each rollout was sampled from Qwen/Qwen2.5-7B-Instruct on
GSM8K test prompts at temperature 0.7.
What's in… See the full description on the dataset page: https://huggingface.co/datasets/eagle0504/multireward-grpo-gsm8k-rewards-qwen2.5-7b.synthetic-text2sql-dataset
Dataset Card for "synthetic-text2sql-dataset"
Dataset Summary
The synthetic-text2sql-dataset is a large-scale, structured dataset containing 100,000 training and 5,851 test examples designed to support research and development in SQL semantic parsing, text-to-SQL generation, and chain-of-thought (CoT) reasoning.
It was derived from an original DataFrame and converted into Hugging Face's datasets.Dataset format. Three new fields were added:
question: alias for the… See the full description on the dataset page: https://huggingface.co/datasets/eagle0504/synthetic-text2sql-dataset.EAGLE3-Apertus-8B-Instruct-2509-Data
EAGLE3-Apertus-8B-Instruct-2509-Data
Training dataset for the thomaskiefer/EAGLE3-Apertus-8B-Instruct-2509 speculative decoding draft model.
Dataset Description
This dataset contains ~375k multi-turn conversations used to train an Eagle3 draft model for swiss-ai/Apertus-8B-Instruct-2509.
Data Sources
The prompts are sourced from:
UltraChat - Large-scale multi-turn dialogue dataset
ShareGPT - Real user conversations
OpenThoughts-114k-math - Mathematical… See the full description on the dataset page: https://huggingface.co/datasets/thomaskiefer/EAGLE3-Apertus-8B-Instruct-2509-Data.MoS-Qwen3-8B-EAGLE3-responses
MoS — Qwen3-8B EAGLE3 Training Responses
Target-model responses for training EAGLE3 speculative-decoding draft models against
Qwen/Qwen3-8B. Built for the MoS (Mixture of
Speculators) project — a routed multi-MLP draft — and equally usable for any single-draft
EAGLE3 / SpecForge training run on Qwen3-8B.
599,087 complete assistant responses (with thinking traces) over five domains, generated
by Qwen3-8B itself so the draft learns to mimic the target's own distribution.… See the full description on the dataset page: https://huggingface.co/datasets/ryan-0608/MoS-Qwen3-8B-EAGLE3-responses.augmented_codealpaca-20k-using-together-ai-deepseek-v1
Dataset Overview
This dataset, named CodeAlpaca-20k, consists of examples that blend coding instructions with outputs and reasoning. Each entry includes structured fields like output, instruction, input, and cot (Chain of Thought). It is particularly designed to train and evaluate AI models that generate code and explanations based on simple programming tasks.
Data Collection and Preparation
Data entries are augmented using the augment_answer function that makes API… See the full description on the dataset page: https://huggingface.co/datasets/eagle0504/augmented_codealpaca-20k-using-together-ai-deepseek-v1.multireward-grpo-fintech-customer-comms
Multi-Reward GRPO — Synthetic Fintech Customer Communications
Synthetic multi-turn customer-service conversations for a fictional bank
("Bank of XYZ"), generated for the empirical Section of "Conditioned
Multi-Reward Advantage Estimation: A Finite-Sample Analysis".
Each conversation ends with m parallel sampled bot replies, each scored
on three verifiable reward channels designed for fintech customer service.
This is the multi-reward GRPO group structure on a real generation… See the full description on the dataset page: https://huggingface.co/datasets/eagle0504/multireward-grpo-fintech-customer-comms.pashto-eagle-1k-cot
Pashto-Eagle-1K-CoT Dataset
Overview
Pashto-Eagle-1K-CoT is a high-fidelity reasoning dataset tailored for the Pashto language. It consists of 1,024 samples featuring complex logic, mathematical reasoning, and step-by-step problem-solving. This dataset is a translated and refined version of the brendan-gho/qwen3b_paraphrased_eagle_cot.
This repository is part of the iPashto.ai initiative to build a robust open-source ecosystem for Pashto Artificial Intelligence, focusing… See the full description on the dataset page: https://huggingface.co/datasets/nassimjp/pashto-eagle-1k-cot.multireward-grpo-gsm8k-rewards
Multi-Reward GRPO — GSM8K Rewards (Qwen2.5-1.5B-Instruct)
Raw rollout-level reward observations from the empirical Section of
"Conditioned Multi-Reward Advantage Estimation: A Finite-Sample Analysis".
This is the data that produced the headline Theorem 3 (correlation-dependent
MSE floor) and Proposition 4 (sign-changing conditioning bias) figures on
real LLM rollouts.
What's in here
For each of 150 GSM8K test prompts, we sampled 16 independent seeds × 32
rollouts… See the full description on the dataset page: https://huggingface.co/datasets/eagle0504/multireward-grpo-gsm8k-rewards.Tobacco-Expert-Dataseteagle3-sarvam-30b-training-data
Eagle3 Sarvam-30B Training Data
Training data used to build the Eagle3 draft model for Sarvam-30B.
Dataset Description
This dataset contains 90,000 prompt-response pairs used to train an Eagle3 speculative decoding draft model for the Sarvam-30B language model.
Each sample consists of a prompt and its corresponding response generated by the Sarvam-30B base model. During training, the model also consumes hidden state features extracted from auxiliary layers of the base… See the full description on the dataset page: https://huggingface.co/datasets/sulabhkatiyar/eagle3-sarvam-30b-training-data.qwen2.5-0.5b-eagle3-dataTobacco-Expert-Dataset2