---
license: mit
language: de
tags:
- news
- headline generation
- teaser generation
- keyword generation
- summarization
- tweet generation
- serp generation
- news snippet generation
size_categories:
- 1K<n<10K
task_categories:
- summarization
- text2text-generation
pretty_name: Instruct-Snippet-MLSUM-500-V2
---

# Dataset Card for Instruct-Snippet-MLSUM-500-V2

### Dataset Summary

This is a dataset for multitask instruction finetuning dataset for the task of news snippet generation. It is built from a sample of ~500 news articles from the [MLSUM](https://huggingface.co/datasets/mlsum) dataset, augmented with machine generated news snippets.

### Supported Tasks

This dataset was created to support the task of generating news snippets such as title, teaser, summary, keywords, serp and tweet for news articles in German language.

### Languages

de - German

## Dataset Structure

lable: a string feature.  
instruction: a string feature.  
input: a string feature.  
output: a string feature.

## Dataset Creation

This dataset was created from Snippet-MLSUM-500-V2. See  [Snippet-MLSUM-500-V2](https://huggingface.co/datasets/snipaid/snippet-mlsum-500-V2) for the dataset without instructions.  
Instructions were generated with GPT-3.5 from a human-curated seed-set of instructions.

## Considerations for Using the Data

### Known Limitations

Part of the snippet data is machine generated. Be aware that these features (specifically: output) may exhibit signs of model hallucination, toxicity and stereotypes.

## Additional Information

### Licensing Information

This dataset is licensed under MIT license.