datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
10k_prompts_ranked
Dataset Card for 10k_prompts_ranked
10k_prompts_ranked is a dataset of prompts with quality rankings created by 314 members of the open-source ML community using Argilla, an open-source tool to label data. The prompts in this dataset include both synthetic and human-generated prompts sourced from a variety of heavily used datasets that include prompts.
The dataset contains 10,331 examples and can be used for training and evaluating language models on prompt ranking tasks. The… See the full description on the dataset page: https://huggingface.co/datasets/data-is-better-together/10k_prompts_ranked.PKU-SafeRLHF-10K
Paper
You can find more information in our paper.
Dataset Paper: https://arxiv.org/abs/2307.04657
China-K12-STEM-10K-CoT-Reasoning
K12-STEM-CoT-Chinese
1.54M Chinese K12 STEM problems with chain-of-thought solutions, 48% with diagrams.
The largest structured Chinese math/physics/chemistry reasoning dataset.
This is a curated sample (10,000 problems) of the full 1.54M dataset available via API.
Full Dataset Access
Access the full 1,540,000+ problems via API →
This Sample
Full API
Total problems
10,025
1,540,000+
With CoT solutions
10,025
1,490,000+
With diagrams
6,093
740,000+… See the full description on the dataset page: https://huggingface.co/datasets/lfaviate/China-K12-STEM-10K-CoT-Reasoning.RAIL-HH-10K RAIL-HH-10K: Multi-Dimensional Safety Alignment Dataset
The first large-scale safety dataset with 99.5% multi-dimensional annotation coverage across 8 ethical dimensions.
📖 Read Blog •
📖 Paper (Coming Soon) •
🚀 Quick Start •
🔌 RAIL API •
💻 Examples
🌟 What Makes RAIL-HH-10K Special?
🎯 Near-Complete Coverage
99.5% dimension coverage across all 8 ethical dimensions
Most existing datasets: 40-70% coverage
RAIL-HH-10K: 98-100%… See the full description on the dataset page: https://huggingface.co/datasets/responsible-ai-labs/RAIL-HH-10K.english-daily-dialogues-10k
English Daily Dialogues 10K
A general-purpose, open dataset of 10,000 synthetic multi-turn English conversations spanning ten everyday-life domains. Built as a clean NLP resource for dialogue modeling, response generation, intent understanding, and conversational evaluation. This is a general language resource — not a safety or security benchmark.
Curated by Enes Deniz (ORCID 0009-0006-9491-3565), Co-Founder at AltaySec. It is the English companion to the Turkish Daily Dialogues… See the full description on the dataset page: https://huggingface.co/datasets/3nesdeniz/english-daily-dialogues-10k.soc-agent-traces-10k
SOC-Agent-Traces-10K
Multi-step SOC investigation agent traces in session-trace format. Each
record is a complete investigation session: an alert arrives, an analyst agent
gathers evidence through nine read-only tools, and closes with a structured
JSON triage report.
Instead of single-turn alert → answer pairs, every record captures the full
reasoning trajectory:
alert → get_surrounding_events → get_process_tree → lookup_attack
→ search_sigma → get_asset_context →… See the full description on the dataset page: https://huggingface.co/datasets/alirezaaminzadeh/soc-agent-traces-10k.telecom-intent-config-sft-10k
Telecom Intent→Config SFT Dataset (10K)
The first open SFT dataset for training LLMs to translate natural language network intents into structured 5G/6G configurations.
This dataset addresses the #1 gap identified in the telecom LLM research landscape: there is no public training dataset for intent-to-policy translation. All existing telecom datasets (TeleQnA, ORANBench-13K, 6G-Bench) are MCQ evaluation benchmarks — not instruction-following format. This dataset fills that gap.… See the full description on the dataset page: https://huggingface.co/datasets/nraptisss/telecom-intent-config-sft-10k.JudgeBias-DPO-RefFree-subset-10k
JudgeBias-DPO-RefFree-subset-10k
A 10K-pair subset of JudgeBias-DPO-RefFree-subset for training LLM judges to evaluate materials science synthesis recipes without bias in a reference-free setting (no ground truth recipe).
Sampling Strategy: Stratified Dataset + Top Delta per Sample
Equal quota per dataset: 9 datasets × ~1,111 pairs = 10,000 total
Within each dataset: for each sample_id, pairs are ranked by score_delta (descending) and selected in round-robin order —… See the full description on the dataset page: https://huggingface.co/datasets/iknow-lab/JudgeBias-DPO-RefFree-subset-10k.ytc-10kNemotron-Math-v2-Medium-10k
Nemotron-Math-v2-Medium-10k
A lightweight 10,500-problem subset of
nvidia/Nemotron-Math-v2
for long-horizon Python-TIR reinforcement learning. It contains 1,500 problems
from each metadata.reason_high_with_tool.pass bucket 1 through 7. A
deterministic seed-42 shuffle assigns 500 examples to validation and 10,000
to train.
This Hugging Face release intentionally contains no teacher traces. The
full messages/tools aggregation is retained as a separate local artifact.… See the full description on the dataset page: https://huggingface.co/datasets/JWei05/Nemotron-Math-v2-Medium-10k.China-K12-STEM-10K-CoT-Reasoning
K12-STEM-CoT-Chinese
1.54M Chinese K12 STEM problems with chain-of-thought solutions, 48% with diagrams.
The largest structured Chinese math/physics/chemistry reasoning dataset.
This is a curated sample (10,000 problems) of the full 1.54M dataset available via API.
Full Dataset Access
Access the full 1,540,000+ problems via API →
This Sample
Full API
Total problems
10,025
1,540,000+
With CoT solutions
10,025
1,490,000+
With diagrams
6,093
740… See the full description on the dataset page: https://huggingface.co/datasets/a13905873166/China-K12-STEM-10K-CoT-Reasoning.prompt-gen-10k-flux-sdxl
Prompt Generation Dataset (10K Narrative for Flux / SDXL)
This dataset (prompt_gen_final_10k.jsonl and prompt_gen_final_10k.csv) was used to train and fine-tune image-prompt models such as KavinduHansaka/Llama-3.2-1B-ImageGen.
It contains 10,000 curated narrative prompt samples designed for image generation models like Stable Diffusion XL and Flux.Unlike raw tag-based datasets, the target field provides natural paragraphs (≈80–100 words) that describe cinematic scenes with… See the full description on the dataset page: https://huggingface.co/datasets/KavinduHansaka/prompt-gen-10k-flux-sdxl.stack-v2-sparse-classes-10k
Stack v2 Sparse Python Classes 10k
This is a 10,000-sample snapshot for Diffusion + Autoregressive hybrid code generation experiments.
Source
The data is extracted from bigcode/the-stack-v2-dedup, Python subset. The extraction uses Stack v2 metadata as source of truth, groups candidates by repo_name + revision_id, fetches files with git partial fetch + sparse checkout, then applies AST-level class filters.
Splits
train.jsonl: 9,000
val.jsonl: 500
test.jsonl:… See the full description on the dataset page: https://huggingface.co/datasets/hybrid-diff-ar/stack-v2-sparse-classes-10k.ru-instruct-10k
10k Russian chatbot dialogues dataset
zelo-scores-10kx100-granite-4.1-30b
Dataset Card for tomaarsen/zelo-scores-10kx100-granite-4.1-30b
Dataset Summary
Synthetic data generated by DataForge:
Model: ibm-granite/granite-4.1-30b (main)
Source dataset: tomaarsen/zelo-pairs-10kx100-quantile-anchor (train split).
Generation config: temperature=None, top_p=None, top_k=None, max_tokens=4096, model_max_context=32768
Speculative decoding: disabled
System prompt: `You are a relevance scoring system. Given a query and two documents (A and B), your job… See the full description on the dataset page: https://huggingface.co/datasets/tomaarsen/zelo-scores-10kx100-granite-4.1-30b.retailopt-10k-verified
RetailOpt-10K Verified
Solver-verified expansion of Jacoblian/RetailOpt-190.
Overview
Property
Value
Base scenarios
190
Variations per base
53
Target total
10,070
Solver
Pyomo + HiGHS
Verification
Optimal objective per scenario
Variation Axes
Each base scenario is perturbed along:
demand_scale — scale demand curves (0.7–1.4×)
capacity_scale — scale production and cold storage capacity (0.6–1.3×)
cost_multiplier — scale… See the full description on the dataset page: https://huggingface.co/datasets/alirezaaminzadeh/retailopt-10k-verified.sp500-edgar-10k-markdown
edgar s&p500
Source Datasets
The source dataset used for this report is jlohding/sp500-edgar-10k.
Dataset Information
Configuration: default
Feature
Data Type
cik
string
sic
string
company
string
date
timestamp[us]
ret
float64
mkt_cap
float64
report_intro
string
text
string
report_returns
string
word_count
int64
Splits:
Train:
Number of Examples: 6258
Size: 2260000389 bytes
Download Size: 974801155 bytesDataset… See the full description on the dataset page: https://huggingface.co/datasets/BEE-spoke-data/sp500-edgar-10k-markdown.Abstract2Appendix_v1_10k
Dataset Card: Abstract2Appendix v1
Dataset Description
The Abstract2Appendix v1 dataset is a high-quality collection of academic peer reviews and their associated research paper metadata. This dataset combines reviews from four premier machine learning and AI conferences: NeurIPS 2023, EMNLP 2023, TMLR, and ICLR 2023, shuffled into a unified corpus. It is designed to enhance long-context capabilities in Large Language Models (LLMs) and supports tasks such as fine-tuning… See the full description on the dataset page: https://huggingface.co/datasets/alexshengzhili/Abstract2Appendix_v1_10k.Nemotron-Math-v2-10k
Nemotron-Math-v2-10k
10,000 machine-verifiable, mid-to-hard math problems subsampled from
nvidia/Nemotron-Math-v2, each with all
stored gpt-oss-120b high-reasoning, no-tool solution traces. Built for RL training
(verifiable final answers) with a teacher-trace column for SFT distillation / analysis.
How it was made (seed 42 throughout)
Dedupe: scanned all 5 splits (high_part00-02, medium, low), deduped rows by
md5(problem) → 324,233 unique problems.… See the full description on the dataset page: https://huggingface.co/datasets/JWei05/Nemotron-Math-v2-10k.code_rose_initial_1_7B_SFT_10K_rollouts_Qwen3-4B-Thinking-2507_k12_t0.7_maxtok12288
code_rose_initial_1_7B_SFT_10K — rollouts (Qwen3-4B-Thinking-2507, k=12)
Pass@k completions generated with vLLM over the prefixes in
CL-From-Nothing/code_rose_initial_1_7B_SFT_10K.
Generation config
Model
Qwen3-4B-Thinking-2507
Samples per question (k)
12
Temperature
0.7
top_p
0.9
max_tokens
12288
max_model_len
32768
Questions
7250 (index 0–7249, full split)
Total rows
87000 (7250 × 12)
Generated by complete_prefix_vllm.py… See the full description on the dataset page: https://huggingface.co/datasets/CL-From-Nothing/code_rose_initial_1_7B_SFT_10K_rollouts_Qwen3-4B-Thinking-2507_k12_t0.7_maxtok12288.alpaca-high-prob-qwen-0.5b-10k
High-Probability Sentence Predictions Dataset
Dataset Description
This dataset contains sentences from tatsu-lab/alpaca
where the model Qwen/Qwen2.5-0.5B predicts the token before the final period
with ≥90% probability.
Source Dataset Attribution
This dataset is derived from tatsu-lab/alpaca
and inherits its license terms (cc-by-nc-4.0). Please cite the original dataset when using this data.
Extraction Parameters
Parameter
Value
Source… See the full description on the dataset page: https://huggingface.co/datasets/ermiaazarkhalili/alpaca-high-prob-qwen-0.5b-10k.pretraining-high-quality-10k-workshop
Lapa HQ 10k Workshop Corpus
A small deterministic subset of lapa-llm/pretraining-high-quality for tokenizer-transfer workshop runs.
Provenance
Source dataset: lapa-llm/pretraining-high-quality
Source config: default
Source split: train
Rows: 10000
Selection: first 10000 rows by dataset-server row order
Download window size: 100
Parallel workers: 20
Created at UTC: 2026-06-20T09:22:21.460168+00:00
Added columns:
source_row_idx
mini_corpus_index
arb-raw-10k
Arb-Agent Raw Data
This is the main text corpus used to train the QuantOxide Reasoning Agent.
It contains clean, semantic text chunks extracted from the 10-K filings of the top 50 S&P 500 companies.
The Parsing Logic
Parsing SEC filings is unbelievably difficult due to inconsistent HTML, broken table tags, and "incorporation by reference." This dataset was created using a unique parsing technique:
Instead of relying on broken regex headers, the parser scores chunks based… See the full description on the dataset page: https://huggingface.co/datasets/ckerf/arb-raw-10k.fable-forge-10k
FableForge — Narrative Reasoning Dataset with Recurrence-Depth Annotations
The first narrative dataset designed around recurrence depth requirements.
Every example carries a suggested_n_loops field with a theoretically grounded basis —
derived from the structural complexity of the task, not a heuristic label or emergent property.
Background
Standard narrative datasets treat reasoning depth as an emergent property. FableForge is
different: it annotates how much… See the full description on the dataset page: https://huggingface.co/datasets/OpenCoven/fable-forge-10k.ocr2_cf1900_k2_qwen35_gpt55_aggregated_10k_seed20260513
Aggregated Qwen3.5 and GPT-5.5 OCR2 Coding Step Labels
This dataset follows the same 500-row parquet layout as JingweiNi/ocr2_cf1900_k2_qwen35_fp8_10k_seed20260513 and aggregates the original Qwen3.5 labels with GPT-5.5 medium reannotation of Qwen3.5-positive error steps.
Aggregation rule: GPT-5.5 confirmed errors are errors; all other 10,000 manifest-selected steps are correct.
Columns
qwen35_verified: original Qwen3.5 labels for the 10,000 manifest-selected steps, NaN… See the full description on the dataset page: https://huggingface.co/datasets/JingweiNi/ocr2_cf1900_k2_qwen35_gpt55_aggregated_10k_seed20260513.alpaca-cleaned-high-prob-qwen-0.5b-10k
High-Probability Sentence Predictions Dataset
Dataset Description
This dataset contains sentences from yahma/alpaca-cleaned
where the model Qwen/Qwen2.5-0.5B predicts the token before the final period
with ≥90% probability.
Source Dataset Attribution
This dataset is derived from yahma/alpaca-cleaned
and inherits its license terms (cc-by-4.0). Please cite the original dataset when using this data.
Extraction Parameters
Parameter
Value… See the full description on the dataset page: https://huggingface.co/datasets/ermiaazarkhalili/alpaca-cleaned-high-prob-qwen-0.5b-10k.alpaca-gpt4-en-high-prob-qwen-0.5b-10k
High-Probability Sentence Predictions Dataset
Dataset Description
This dataset contains sentences from llamafactory/alpaca_gpt4_en
where the model Qwen/Qwen2.5-0.5B predicts the token before the final period
with ≥90% probability.
Source Dataset Attribution
This dataset is derived from llamafactory/alpaca_gpt4_en
and inherits its license terms (apache-2.0). Please cite the original dataset when using this data.
Extraction Parameters
Parameter… See the full description on the dataset page: https://huggingface.co/datasets/ermiaazarkhalili/alpaca-gpt4-en-high-prob-qwen-0.5b-10k.orca-mini-v1-high-prob-qwen-0.5b-10k
High-Probability Sentence Predictions Dataset
Dataset Description
This dataset contains sentences from pankajmathur/orca_mini_v1_dataset
where the model Qwen/Qwen2.5-0.5B predicts the token before the final period
with ≥90% probability.
Source Dataset Attribution
This dataset is derived from pankajmathur/orca_mini_v1_dataset
and inherits its license terms (apache-2.0). Please cite the original dataset when using this data.
Extraction Parameters… See the full description on the dataset page: https://huggingface.co/datasets/ermiaazarkhalili/orca-mini-v1-high-prob-qwen-0.5b-10k.GHIA-CHRONOS-Synthetic-Dialogue-10K
🌌 GHIA-CHRONOS: The Industrial Ops Corpus
A Recursive Civilization Simulation Dataset for Long-Horizon AI Reasoning
📘 Dataset Overview
Field
Information
Dataset Name
GHIA-CHRONOS
Dataset Type
Synthetic Recursive Civilization Dataset
Primary Purpose
Long-horizon reasoning, relativistic causality, strategic simulation
Data Format
JSONL
Generation Style
Optimized low-power recursive streaming
Current Public Sample
10,000 records
Master… See the full description on the dataset page: https://huggingface.co/datasets/Sangadi-Bujji/GHIA-CHRONOS-Synthetic-Dialogue-10K.sinhala-validation-set-10k
Sinhala Validation Set - 10K Sentences
A held-out Sinhala validation set of 10,000 sentences drawn from the Minuri/diverse_sinhala_dataset corpus. Used for monitoring validation loss during continual pretraining of three LLaMA 3.2 1B variants (Models A, B, C) as part of a diversity-driven Sinhala language model adaptation study.
Dataset Description
This validation set was held out strictly from all three pretraining corpora (A, B, C) to enable unbiased validation loss… See the full description on the dataset page: https://huggingface.co/datasets/Minuri/sinhala-validation-set-10k.