2A2I/H4_no_robots

Dataset Card for "No Robots" 🙅‍♂️🤖

Summary

"No Robots" is a dataset consisting of 10,000 instructions and demonstrations, created by professional annotators. It was translated using the Google Cloud Platform Translation API. This dataset can be used to train language models to follow instructions more accurately (instruction-tuned fine-tuning - SFT). The "No Robots" dataset was created based on the dataset described in OpenAI's InstructGPT paper, and includes the following categories:

Category	Count
Creation	4560
Open Questions	1240
Brainstorming	1120
Chatting	850
Rewriting	660
Summarization	420
Programming	350
Classification	350
Closed Questions	260
Extraction	190

Languages

This dataset is available in Arabic only. The original version in English can be found at this link, and the Turkish version at this link.

Data Fields

Columns as follows:

• prompt: Specifies the instruction that the model should follow.
• prompt_id: A unique identifier.
• messages: A list containing dictionaries, each dictionary describes a message (key: content) and who sent it (key: role).
• category: The task category, I did not translate this.

Splits

	train	test
No Robots	9500	500

License

The dataset is available under the (CC BY-NC 4.0) license.

Citation Information

@misc{no_robots,
  author = {Nazneen Rajani and Lewis Tunstall and Edward Beeching and Nathan Lambert and Alexander M. Rush and Thomas Wolf},
  title = {No Robots},
  year = {2023},
  publisher = {Hugging Face},
  journal = {Hugging Face repository},
  howpublished = {\url{https://huggingface.co/datasets/HuggingFaceH4/no_robots}}
}