metadata
language:
- en
license: other
task_categories:
- question-answering
- text-generation
tags:
- question-answering
- text-generation
- chat
- instruction-tuning
- non-synthetic
- dl26
size_categories:
- 100K<n<1M
UltraAtlas
UltraAtlas is a high-quality English question-answering dataset prepared by Dl26 for text-generation and chat-style model training.
The dataset is converted from SQuAD v2 into a consistent assistant format with prompt, response, and messages fields. It is intended for supervised fine-tuning of general assistant models that need strong question answering behavior.
Dataset Details
| Property | Value |
|---|---|
| Dataset name | UltraAtlas |
| Developer | Dl26 |
| Source dataset | rajpurkar/squad_v2 |
| Rows | 130,319 |
| Language | English |
| Task | Question answering / text generation |
| Format | Single-turn chat with user and assistant messages |
| Synthetic data | No model-generated synthetic answers are added |
Fields
| Field | Description |
|---|---|
id |
Stable row identifier for this converted dataset |
source_dataset |
Upstream dataset identifier |
source_id |
Upstream row or question identifier when available |
task |
QA task family |
prompt |
User-facing instruction/question text |
response |
Ground-truth answer text |
messages |
Chat-format list with user and assistant turns |
quality_source |
Short provenance note |
Example
from datasets import load_dataset
dataset = load_dataset("Dl26/UltraAtlas", split="train")
print(dataset[0]["messages"])
Intended Use
UltraAtlas is intended for:
- supervised fine-tuning of text-generation models
- question-answering behavior training
- single-turn assistant training
- retrieval and answer-grounding experiments
- general English QA evaluation and data mixing
Source and Licensing
This dataset is a format conversion of rajpurkar/squad_v2. The conversion keeps source provenance fields so users can inspect origin and terms. Upstream dataset licenses and terms still apply.