datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
full-structured-instruction-sft-dataset
Full Structured + Instruction SFT Corpus
Unified SFT training corpus built from Glaive, Hermes, UltraChat, and synthetic structured-output data.
Dataset repo
mdonigian/full-structured-instruction-sft-datasetRelease date: 2026-03-11
Included files
train_full_sft.jsonl: full merged and shuffled SFT dataset
source_glaive.jsonl: processed Glaive subset
source_hermes.jsonl: processed Hermes subset
source_ultrachat.jsonl: processed UltraChat subset… See the full description on the dataset page: https://huggingface.co/datasets/mdonigian/full-structured-instruction-sft-dataset.sft-tool-calling-structured-output-v1
vericava/sft-tool-calling-structured-output-v1
Dataset to train (SFT) 3-20B LLMs for tool calling and structured outputs/classifications.
Includes contents in English as well as some Japanese.
structured-output-sft-100k
Structured Output SFT (100K)
100,000 ShareGPT conversations demonstrating correct generation of structured data formats: JSON, YAML, CSV, XML, Markdown tables, JSON Schema, and OpenAPI fragments. Each example pairs a natural language specification with a valid, well-formed output.
Motivation
Structured output generation is among the most commercially critical LLM capabilities. Models fail in characteristic ways:
Invalid JSON: unclosed brackets, trailing commas… See the full description on the dataset page: https://huggingface.co/datasets/stindardlogic/structured-output-sft-100k.enwiki_structured_content
Dataset Card for enwiki_structured_content
Dataset Description
This dataset is derived from the early official Wikipedia release, downloaded from the en subset of Wikipedia Structured Contents.Articles were converted to Markdown.
structured-hard-sft-4k
Hard Synthetic Dataset for Structured Data Tasks (v1)
This dataset contains 4,000 high-difficulty synthetic samples designed to improve LLM performance on complex structured data conversion, extraction, and formatting tasks.
The data is fully synthetic, generated using deterministic serialization to ensure syntax validity while maintaining high structural complexity (deep nesting and varied types).
Dataset Summary
The dataset addresses four "hard" areas typically… See the full description on the dataset page: https://huggingface.co/datasets/daichira/structured-hard-sft-4k.structured_medicalThe dataset was presented in the paper Gazal-R1: Achieving State-of-the-Art Medical Reasoning with Parameter-Efficient Two-Stage Training.
structured-hard-sft-4k
Hard Synthetic Dataset for Structured Data Tasks (v1)
This dataset contains 4,000 high-difficulty synthetic samples designed to improve LLM performance on complex structured data conversion, extraction, and formatting tasks.
The data is fully synthetic, generated using deterministic serialization to ensure syntax validity while maintaining high structural complexity (deep nesting and varied types).
Dataset Summary
The dataset addresses four "hard" areas typically… See the full description on the dataset page: https://huggingface.co/datasets/zyz123code/structured-hard-sft-4k.toricblm-dataset-state-toricblm-structure-priority-balanced-3day-20260709t185034z-epoch-001
ToricBLM dataset state: toricblm-structure-priority-balanced-3day-20260709T185034Z epoch 001
This dataset repo records the exact local training-data state visible to the dynamic epoch launcher.
It intentionally stores manifests and audit records rather than duplicating large Parquet shards.
Special checkpoint: toricblm-structure-priority-balanced-3day-20260709T185034Z_epoch_001_special_structure_current_step_002000.pt
Checkpoint repo: AmelieSchreiber/ToricGT_160M_FoT
Curriculum… See the full description on the dataset page: https://huggingface.co/datasets/AmelieSchreiber/toricblm-dataset-state-toricblm-structure-priority-balanced-3day-20260709t185034z-epoch-001.toricblm-dataset-state-toricblm-structure-priority-balanced-3day-20260709t185034z-epoch-002
ToricBLM dataset state: toricblm-structure-priority-balanced-3day-20260709T185034Z epoch 002
This dataset repo records the exact local training-data state visible to the dynamic epoch launcher.
It intentionally stores manifests and audit records rather than duplicating large Parquet shards.
Special checkpoint: toricblm-structure-priority-balanced-3day-20260709T185034Z_epoch_002_special_structure_delta_step_002750.pt
Checkpoint repo: AmelieSchreiber/ToricGT_160M_FoT
Curriculum… See the full description on the dataset page: https://huggingface.co/datasets/AmelieSchreiber/toricblm-dataset-state-toricblm-structure-priority-balanced-3day-20260709t185034z-epoch-002.structured-5k-mix-sft
5k Mixed Hard-Structured SFT Dataset (v1)
This dataset contains 5,000 synthetic samples designed to improve LLM performance on complex structured data conversion, extraction, and formatting tasks.
It aggregates 13 distinct conversion tasks with a specific focus on format diversity and structural complexity.
Dataset Summary
The dataset is distributed across five major formats with the following allocation:
Target Format
Count
Share
Task Types
YAML
1,500
30%… See the full description on the dataset page: https://huggingface.co/datasets/daichira/structured-5k-mix-sft.structured-stern-neon-articles
Structured Stern NEON Community Articles
This repository contains approximately 20k user written texts,
articles, and poetry pulled from archives of the Stern NEON website.
Stern NEON was a community platform where users could write and publish their own articles.
Many of the articles are personal stories, poems, or opinion pieces.
The articles are structured in a way that they can be used for further analysis.
Dataset Details
Uses
This dataset can be used for… See the full description on the dataset page: https://huggingface.co/datasets/dotwee/structured-stern-neon-articles.stage3-synthetic-structured-retrieval
Stage 3 Synthetic Structured-Retrieval Agents
Native search-tool trajectories generated by
Qwen/Qwen3-235B-A22B-Instruct-2507 for LCLM Stage-3 agent post-training.
The default config contains only traces that passed programmatic evidence and
answer verification.
Harvest
Accepted traces: 82
Native search calls: 179
Compressed tool-observation traces: 41
Uncompressed traces: 41
Task-ID overlap between pilot and collection batch: 0
Family counts are 20 latest-state… See the full description on the dataset page: https://huggingface.co/datasets/leonli66/stage3-synthetic-structured-retrieval.json-structured-output-dpo-3k
JSON Structured Output DPO Pairs (3K)
DPO preference pairs for training LLMs to produce valid, schema-compliant JSON output.
Motivation
Structured output (JSON mode) is critical for production AI applications — parsers fail, pipelines break, and downstream processing errors when models output malformed JSON, use wrong field names, or wrap responses in markdown. This dataset trains strict schema adherence.
Dataset Description
3,000 preference pairs… See the full description on the dataset page: https://huggingface.co/datasets/stindardlogic/json-structured-output-dpo-3k.Structured-Todo-Lists-for-Learning-and-Projects
Academic Task Management Dataset
Overview
100 structured todo lists for academic and personal organization. Culturally diverse with 70% Indian education context, 25% European scenarios, and 5% other Asian contexts.
Dataset Structure
{
"input": "Task description",
"output": {
"type": "todo",
"title": "List title",
"category": "academic/personal/project",
"items": [
{"task": "...", "done": false, "priority": "low/medium/high"}
]
}… See the full description on the dataset page: https://huggingface.co/datasets/Srinivasmec26/Structured-Todo-Lists-for-Learning-and-Projects.structured-3k-mix-sft
3k Mixed Hard-Structured SFT Dataset (v1)
This dataset contains 3,000 synthetic samples designed to improve LLM performance on complex structured data conversion, extraction, and formatting tasks.
It features a perfectly balanced distribution across 5 major data formats, ensuring comprehensive coverage of structural logic.
Dataset Summary
The dataset consists of 3,000 samples, with an equal allocation of 20% (600 samples) for each target format.
Target Format
Count… See the full description on the dataset page: https://huggingface.co/datasets/daichira/structured-3k-mix-sft.structured_medicalThe dataset was presented in the paper Gazal-R1: Achieving State-of-the-Art Medical Reasoning with Parameter-Efficient Two-Stage Training.
structured-3k-mix-sft-orpo
3k Mixed Hard-Structured ORPO Dataset (v2)
更新概要 (v1 → v2)
項目
v1
v2
総ペア数
2,294
2,041
主なルール
C_EXTRA_TEXT (27%), C_CODE_FENCE (19%) など表層的なもの
内容レベルの破壊のみ
rewards/accuracies 問題
0.9625（trivial すぎた）
コンテンツ変形で改善
max_total_chars フィルタ
3072 (XMLほぼ除外)
8000 (5形式均等)
データセット概要
構造化データ生成タスク（JSON/YAML/XML/TOML/CSV）に特化した ORPO (Odds Ratio Preference Optimization) 学習用データセットです。
ベースデータ: daichira/structured-3k-mix-sft
v2 のルール方針
v2 の最大の変更点:… See the full description on the dataset page: https://huggingface.co/datasets/daichira/structured-3k-mix-sft-orpo.