datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Chinese-High-School-Chemistry-Correction-Dataset
Chinese-High-School-Chemistry-Correction-Dataset
一个面向「高中化学垂直大模型微调」的中文问答与文本生成数据集
1. 数据集缘起
为了训练一个高中化学领域的垂直大模型，我们需要大量高质量、结构化的中文语料。本数据集整理了三版主流教科书、常考化学方程式与畅销教辅等中的知识点，全部转为统一的 JSONL 格式。
2. 数据来源
普通高中教科书（苏教版、人教版、鲁教版）、高中常考化学方程式、高中参考教辅资料（一本涂书、教材帮等）均转成jsonl格式
该jsonl文件数据，部分行或许有格式错误，需要自行编写py脚本校对，以便用于大模型微调。
3. 数据格式（JSONL）
每行一条记录，可直接用于 Hugging Face datasets 库：
{"instruction": "已知0.5 mol的水（H₂O）的质量是9 g，且含有3.01×10²³个水分子。请计算1 mol水的质量和阿伏伽德罗常数。", "output":… See the full description on the dataset page: https://huggingface.co/datasets/liushuaiqian/Chinese-High-School-Chemistry-Correction-Dataset.sanskrit-ocr-post-correction\
A Benchmark and Dataset for Post-OCR text correction in Sanskrit.
This dataset contains manually post-edited OCR data for Sanskrit texts in Devanagari script.
It includes:
- Train/Validation/Test splits with OCR text and corrected ground truth
- An out-of-domain test set of 500 sentences
- Source texts from classical Sanskrit works including Brahmasutra Bhashyam, Grahalaghava, and Goladhyayayoutube_caption_corrections
Dataset Card for YouTube Caption Corrections
Dataset Summary
This dataset is built from pairs of YouTube captions where both an auto-generated and a manually-corrected caption are available for a single specified language. It currently only in English, but scripts at repo support other languages. The motivation for creating it was from viewing errors in auto-generated captions at a recent virtual conference, with the hope that there could be some way to help correct those… See the full description on the dataset page: https://huggingface.co/datasets/community-datasets/youtube_caption_corrections.task590_amazonfood_summary_correction_classification
Dataset Card for Natural Instructions (https://github.com/allenai/natural-instructions) Task: task590_amazonfood_summary_correction_classification
Additional Information
Citation Information
The following paper introduces the corpus in detail. If you use the corpus in published work, please cite it:
@misc{wang2022supernaturalinstructionsgeneralizationdeclarativeinstructions,
title={Super-NaturalInstructions: Generalization via Declarative Instructions on… See the full description on the dataset page: https://huggingface.co/datasets/Lots-of-LoRAs/task590_amazonfood_summary_correction_classification.fable-5-coding-and-debugging-traces-synthetic-corrections
Model Synthetic Corrections
1 TRAJECTORIES · 2 TRAINING ROWS · 16 kB
Generated by moonshiner — an open harness for
distilling verified instruction-following, tool-use, and agentic coding traces.
Synthetic Corrections companion dataset. The original dataset is greghavens/fable-5-coding-and-debugging-traces. These are narrowly, synthetically corrected, independently re-judged traces that never passed in the original dataset.
Behavior-preserving instruction-following… See the full description on the dataset page: https://huggingface.co/datasets/greghavens/fable-5-coding-and-debugging-traces-synthetic-corrections.vsec-vietnamese-spell-correction
VSEC: Vietnamese Spell Correction Dataset
Dataset Description
VSEC (Vietnamese Spell Correction) is a comprehensive dataset for Vietnamese spelling error detection and correction, containing 9,341 sentences with 11,202 human-made misspellings across 5,211 unique error types. This dataset represents the largest publicly available collection of Vietnamese spelling errors with syllable-level annotations, making it an invaluable resource for developing and evaluating… See the full description on the dataset page: https://huggingface.co/datasets/nguyenthanhasia/vsec-vietnamese-spell-correction.Indic-post-ocr-correction
Indic Contextual Post-OCR Correction
Dataset Summary
This dataset supports contextual post-OCR correction for Indic languages. Each example is a sentence-level triple consisting of:
an OCR-generated sentence (noisy),
the preceding sentence used as context, and
the corrected sentence (ground truth).
Hugging Face dataset page:
https://huggingface.co/datasets/AbhishekBhandari/Indic-post-ocr-correction
Supported Tasks
Post-OCR text correction… See the full description on the dataset page: https://huggingface.co/datasets/AbhishekBhandari/Indic-post-ocr-correction.task587_amazonfood_polarity_correction_classification
Dataset Card for Natural Instructions (https://github.com/allenai/natural-instructions) Task: task587_amazonfood_polarity_correction_classification
Additional Information
Citation Information
The following paper introduces the corpus in detail. If you use the corpus in published work, please cite it:
@misc{wang2022supernaturalinstructionsgeneralizationdeclarativeinstructions,
title={Super-NaturalInstructions: Generalization via Declarative Instructions on… See the full description on the dataset page: https://huggingface.co/datasets/Lots-of-LoRAs/task587_amazonfood_polarity_correction_classification.vn-spell-correction-eval-real
vn-spell-correction-eval-real
Out-of-distribution evaluation corpus for Vietnamese spell-correction
models — 150 hand-curated (noisy, clean) pairs sampled from real
VN error sources, not generated by nom.text.noise.
This is the test set we use to verify a spell-correction model
generalises beyond its own synthetic training distribution. A model
that scores 95 % on nom-vn's synthetic eval grid and 60 % on this
set is overfit to the noise generator.
Splits
Config… See the full description on the dataset page: https://huggingface.co/datasets/nrl-ai/vn-spell-correction-eval-real.vi-spelling-correction
Vietnamese Spelling Correction Dataset
This dataset contains 978,417 pairs of noisy (source) and clean (target) Vietnamese sentences, designed for training spelling correction models.
The dataset was synthetically generated by injecting realistic noise into a clean Vietnamese corpus.
Dataset Structure
The dataset is divided into training and testing sets:
Train: 880,575 examples
Test: 97,842 examples
Data Fields
source: The text with injected errors (input).… See the full description on the dataset page: https://huggingface.co/datasets/coung21/vi-spelling-correction.tibetan-spelling-correction-dataset
Tibetan Spelling Correction
Sentence-level spelling correction pairs for Tibetan. Each row pairs an annotator's transcription of a manuscript page segment with the final reviewer's corrected version, from the double-annotation workflow of the BDRC Etext Corpus.
For training and evaluating post-correction models such as TiSpell.
Source batches: Ume 1-4, Uchen 1-4. 4,672 pages.
Contents
train
validation
test
All
error pairs
43,015
2,268
2,229
47,512… See the full description on the dataset page: https://huggingface.co/datasets/BDRC/tibetan-spelling-correction-dataset.bias-correction-palestine-protocol
Dataset Card for LLM Bias Correction (Palestine/Israel Context)
This dataset is an open-source alignment and alignment-tuning asset configured explicitly to counteract systemic institutional bias, false symmetry ("both-sidesism"), and documented data manipulation layers regarding the material realities of Palestine and Israel.
Dataset Structure
The asset uses a three-field structure that can be transformed for Supervised Fine-Tuning (SFT) or preference-training… See the full description on the dataset page: https://huggingface.co/datasets/tech-equity-collective/bias-correction-palestine-protocol.spell-correction-ru
Spell Correction RU — датасеты для коррекции ошибок в русском тексте
Набор данных для обучения моделей исправления орфографических, пунктуационных и регистровых ошибок в русскоязычных текстах.
Каждый пример — пара «правильный текст» → «текст с ошибкой». Датасет использовался для обучения модели melsmm/Spell-Corrector-RU-4B.
📦 Код генерации, ноутбуки и полное описание проекта: github.com/melsmm/llm-spell-corrector
Состав
Датасет содержит две конфигурации… See the full description on the dataset page: https://huggingface.co/datasets/melsmm/spell-correction-ru.gemma4-onpolicy-student-corrections
Gemma 4 12B FrontierDistill - On-Policy Student Failure Corrections
Attribution Requirement: This dataset was created and curated by True2456. Any use, redistribution, derivative dataset, model fine-tune, or paper using this dataset MUST cite and reference True2456 and the Gemma 4 12B FrontierDistill Project.
This dataset contains 2,000 on-policy student failure corrections collected live from Gemma 4 12B (gemma-4-12b-it-qat-frontierdistill).
Every example in this dataset… See the full description on the dataset page: https://huggingface.co/datasets/True2456/gemma4-onpolicy-student-corrections.synthetic-self-correction-and-thinking-samples
Self Correction and Thinking
A seed library for training language models to reason with self-correction.
Teaches three reasoning behaviors -- catching your own errors, verifying correct answers, and rejecting false doubts -- across four domains, three difficulty tiers, and three reasoning modes. Also includes multi-turn user-correction conversations where the user actively corrects or challenges the assistant.
The structure at a glance
graph TB… See the full description on the dataset page: https://huggingface.co/datasets/sbussiso/synthetic-self-correction-and-thinking-samples.vn-text-correction-0001
Dataset Card for Vietnamese Text Correction Dataset
Dataset Description
This dataset contains Vietnamese text pairs for training and evaluating text correction models. Each example consists of an erroneous text and its corrected version, making it ideal for:
Grammar correction
Spelling correction
Text normalization
Language model fine-tuning
Dataset Summary
Language: Vietnamese (vi)
Format: Text correction pairs
Size: ~4.0M examples across… See the full description on the dataset page: https://huggingface.co/datasets/LakoreAI/vn-text-correction-0001.ko-finance-asr-corrections
ko-finance-asr-corrections
Frequency-annotated Korean ASR confusion pairs from finance/stock YouTube.
210 pairs
mined from 2,391 videos of auto-captions
across 47 channels
totalling 1,080.1 hours
Each pair carries how often the term was mangled and how often it was said correctly, plus
verification provenance.
한국어 금융·주식 유튜브 자동자막에서 실측한 ASR 오인식→교정 쌍입니다. 모든 쌍에 오표기·정답
표기 빈도(→ 용어별 오인식률)와 검증 메타데이터(2-LLM 합의 감사, 승격 티어)가 붙어 있습니다.
What makes it different
No public… See the full description on the dataset page: https://huggingface.co/datasets/woongstar/ko-finance-asr-corrections.python-runtime-verified-error-correction
Python Runtime-Verified Error Correction Dataset 🐍⚡
Overview
Production-grade synthetic dataset of Python code errors with runtime-verified corrections. Each sample contains broken code, the actual runtime error, and a guaranteed-working fix validated through execution.
Unlike traditional synthetic datasets, every correction is verified by actually running the code in an isolated environment—eliminating hallucinations and ensuring real-world applicability.… See the full description on the dataset page: https://huggingface.co/datasets/SyntheticLogic-Labs/python-runtime-verified-error-correction.text-correction_collection
Human Samples
These samples contains contains human-written sentences produced during language learning practice, combined with AI-based grammatical verification and correction. The original sentences were written by language learners who often did not know whether their sentences were correct or incorrect. These authentic learner inputs capture a wide range of natural mistakes, such as spelling, syntax, word choice, and structure errors.
Synthetic Samples
These… See the full description on the dataset page: https://huggingface.co/datasets/marcelone/text-correction_collection.gemma4-onpolicy-50topics-2000-corrections
Gemma 4 12B FrontierDistill - 2,000 Authentic 50-Topics On-Policy Student Failure Corrections
Attribution Requirement: This dataset was created and curated by True2456. Any use, redistribution, derivative dataset, model fine-tune, or paper using this dataset MUST cite and reference True2456 and the Gemma 4 12B FrontierDistill Project.
This dataset contains 2,000 authentic on-policy student failure corrections collected live from Gemma 4 12B (gemma-4-12b-it-qat-frontierdistill)… See the full description on the dataset page: https://huggingface.co/datasets/True2456/gemma4-onpolicy-50topics-2000-corrections.ocr-correction
OCR (Optical Character Recognition) Correction Dataset
This dataset comprises OCR-corrected text samples from English books and newspapers sourced from the Internet Archive. It provides pairs of raw OCR text and their AI-corrected versions, designed for OCR correction tasks.
Dataset Structure
Data Instances
Each instance contains:
input: Raw OCR text with errors
output: Corrected text
Example:
{
"input": "\n\n(ii) The income of Tarai and Bhabar… See the full description on the dataset page: https://huggingface.co/datasets/agentlans/ocr-correction.sentence-correction
LLM Prompt Dataset
Overview
The LLM Prompt Dataset is designed to enhance the performance of large language models (LLMs) by transforming user inputs into structured prompts. This dataset aims to facilitate the understanding of complex queries and improve the interaction between users and LLMs.
Dataset Structure
The dataset is organized in JSON format, where each entry consists of an input and a prompt. The input represents the original user query or statement… See the full description on the dataset page: https://huggingface.co/datasets/Ashraf-CK/sentence-correction.vn-spell-correction-train
nrl-ai/vn-spell-correction-train
459,478 (noisy, clean) Vietnamese training pairs for fine-tuning a
seq2seq spell-correction model. Each row:
{"input": "<noisy>", "target": "<clean>"}
Both fields are NFC-normalized.
How it was built
Clean side: same 500K register-balanced mix as
nrl-ai/vn-diacritic-train —
350K Vietnamese Wikipedia (CC-BY-SA-4.0,
hirine/wikipedia-vietnamese-1M296K-dataset) + 150K NFC-fixed
Vietnamese news (CC-BY-4.0, tmnam20/Vietnamese-News-dedup).… See the full description on the dataset page: https://huggingface.co/datasets/nrl-ai/vn-spell-correction-train.pleias-post-ocr-correction-chonkie-aligned-en
PleIAs Post-OCR Correction — Chonkie-Aligned Semantic Chunks
This dataset is a semantically chunked and span-aligned derivative of PleIAs/Post-OCR-Correction.
Each record contains:
an OCR hypothesis chunk from the original text field;
a corresponding post-OCR correction output chunk from the corrected_text field;
metadata inherited from the PleIAs dataset;
character spans linking each chunk back to the original source document;
alignment diagnostics produced during filtering.… See the full description on the dataset page: https://huggingface.co/datasets/emanuelaboros/pleias-post-ocr-correction-chonkie-aligned-en.gec-targeted-corrections-esl
GEC Targeted Corrections — ESL
An LLM-generated grammatical error correction (GEC) dataset targeting the specific error patterns that ESL learners most commonly produce. Each example is a (src, tgt) pair where src contains a realistic grammatical error and tgt is the minimally corrected version: only what is necessary is changed.
Dataset Summary
Split
Examples
train
2,037
Schema
{
"src": "She gave me some advices about the… See the full description on the dataset page: https://huggingface.co/datasets/martinsr/gec-targeted-corrections-esl.khmer-spelling-corrections
Khmer Spelling Corrections
Naturally occurring Khmer misspellings paired with the word the writer meant.
The labels are not annotated, they are observed. Search sessions record the
whole typing trajectory toward a single word, so when a user types something,
fails, adjusts and lands on a real dictionary headword, the failed attempt and
the headword form a correction pair produced by a real person under no
instruction to make mistakes.
Only pairs within two edits of the target… See the full description on the dataset page: https://huggingface.co/datasets/seanghay/khmer-spelling-corrections.bf-vibe-bibframe-corrections
BF Vibe BIBFRAME Corrections (v3)
Training data for the single open model behind BF Vibe, a desktop
assistant that helps catalogers create and repair valid BIBFRAME RDF/XML.
The core task is BIBFRAME correction — (corrupted → conforming) record
pairs verified against SHACL shapes — supplemented by two smaller task
families that teach the model to operate the application: routing free-text
requests to BF Vibe's commands and calling its grounding tools.
Created by: Jim Hahn… See the full description on the dataset page: https://huggingface.co/datasets/jimfhahn/bf-vibe-bibframe-corrections.vn-spell-correction-eval
nrl-ai/vn-spell-correction-eval
Vietnamese spell-correction evaluation grid: 4 source registers × 2
noise levels = 8 splits, 2,098 (noisy, clean) sentence pairs total.
Each pair is {"input": "<noisy>", "target": "<clean>"}. Both sides are
NFC-normalized. The clean target is the same sentence used as the
target in nrl-ai/vn-diacritic-eval —
spell correction is a strict superset of diacritic restoration, so we
reuse the same registers-balanced corpus.
Splits
Two noise… See the full description on the dataset page: https://huggingface.co/datasets/nrl-ai/vn-spell-correction-eval.babble-corrections
babble — corrections
Training data for babble: a ~3M parameter
byte-level transformer that started from random weights and has only ever
learned from people correcting it in Discord.
There is no pretraining corpus. There is no scraped chat history. Every row here
is somebody deliberately teaching a small confused model to talk.
How a row happens
Someone @mentions the bot.
The bot replies with whatever its current weights produce. Early on this is
noise, and it is… See the full description on the dataset page: https://huggingface.co/datasets/kowo-co/babble-corrections.thai-misspell-correction
Thai Misspell Correction
This dataset was create from VISTEC-TP-TH-21 dataset. It can use for training Thai Misspell correction model or LLM model.
Use case: Thai Misspell correction
inputs: Text
targets: Text correction
Domain: social media
VISTEC-TP-TH-21 dataset: https://github.com/mrpeerat/OSKut/tree/main/VISTEC-TP-TH-2021
Citation
@inproceedings{limkonchotiwat-etal-2021-handling,
title = "Handling Cross- and Out-of-Domain Samples in {T}hai Word… See the full description on the dataset page: https://huggingface.co/datasets/pythainlp/thai-misspell-correction.