Dataset Card for Alpaca

Dataset Summary

Alpaca is a dataset of 52,000 instructions and demonstrations generated by OpenAI's text-davinci-003 engine. This instruction data can be used to conduct instruction-tuning for language models and make the language model follow instruction better.

The authors built on the data generation pipeline from Self-Instruct framework and made the following modifications:

• The text-davinci-003 engine to generate the instruction data instead of davinci.
• A new prompt was written that explicitly gave the requirement of instruction generation to text-davinci-003.
• Much more aggressive batch decoding was used, i.e., generating 20 instructions at once, which significantly reduced the cost of data generation.
• The data generation pipeline was simplified by discarding the difference between classification and non-classification instructions.
• Only a single instance was generated for each instruction, instead of 2 to 3 instances as in Self-Instruct.

This produced an instruction-following dataset with 52K examples obtained at a much lower cost (less than $500). In a preliminary study, the authors also found that the 52K generated data to be much more diverse than the data released by Self-Instruct.

Supported Tasks and Leaderboards

The Alpaca dataset designed for instruction training pretrained language models.

Languages

The data in Alpaca are in English (BCP-47 en).

Dataset Structure

Data Instances

An example of "train" looks as follows:

{
    "instruction": "Create a classification task by clustering the given list of items.",
    "input": "Apples, oranges, bananas, strawberries, pineapples",
    "output": "Class 1: Apples, Oranges\nClass 2: Bananas, Strawberries\nClass 3: Pineapples",
    "text": "Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.\n\n### Instruction:\nCreate a classification task by clustering the given list of items.\n\n### Input:\nApples, oranges, bananas, strawberries, pineapples\n\n### Response:\nClass 1: Apples, Oranges\nClass 2: Bananas, Strawberries\nClass 3: Pineapples",
}

Data Fields

The data fields are as follows:

• instruction: describes the task the model should perform. Each of the 52K instructions is unique.
• input: optional context or input for the task. For example, when the instruction is "Summarize the following article", the input is the article. Around 40% of the examples have an input.
• output: the answer to the instruction as generated by text-davinci-003.
• text: the instruction, input and output formatted with the prompt template used by the authors for fine-tuning their models.

Data Splits

Dataset Creation

Curation Rationale

[More Information Needed]

Source Data

Initial Data Collection and Normalization

[More Information Needed]

Who are the source language producers?

[More Information Needed]

Annotations

Annotation process

[More Information Needed]

Who are the annotators?

[More Information Needed]

Personal and Sensitive Information

[More Information Needed]

Considerations for Using the Data

Social Impact of Dataset

Excerpt the blog post accompanying the release of this dataset:

We believe that releasing the above assets will enable the academic community to perform controlled scientific studies on instruction-following language models, resulting in better science and ultimately new techniques to address the existing deficiencies with these models. At the same time, any release carries some risk. First, we recognize that releasing our training recipe reveals the feasibility of certain capabilities. On one hand, this enables more people (including bad actors) to create models that could cause harm (either intentionally or not). On the other hand, this awareness might incentivize swift defensive action, especially from the academic community, now empowered by the means to perform deeper safety research on such models. Overall, we believe that the benefits for the research community outweigh the risks of this particular release. Given that we are releasing the training recipe, we believe that releasing the data, model weights, and training code incur minimal further risk, given the simplicity of the recipe. At the same time, releasing these assets has enormous benefits for reproducible science, so that the academic community can use standard datasets, models, and code to perform controlled comparisons and to explore extensions. Deploying an interactive demo for Alpaca also poses potential risks, such as more widely disseminating harmful content and lowering the barrier for spam, fraud, or disinformation. We have put into place two risk mitigation strategies. First, we have implemented a content filter using OpenAI’s content moderation API, which filters out harmful content as defined by OpenAI’s usage policies. Second, we watermark all the model outputs using the method described in Kirchenbauer et al. 2023, so that others can detect (with some probability) whether an output comes from Alpaca 7B. Finally, we have strict terms and conditions for using the demo; it is restricted to non-commercial uses and to uses that follow LLaMA’s license agreement. We understand that these mitigation measures can be circumvented once we release the model weights or if users train their own instruction-following models. However, by installing these mitigations, we hope to advance the best practices and ultimately develop community norms for the responsible deployment of foundation models.

Discussion of Biases

[More Information Needed]

Other Known Limitations

The alpaca data is generated by a language model (text-davinci-003) and inevitably contains some errors or biases. We encourage users to use this data with caution and propose new methods to filter or improve the imperfections.

Additional Information

Dataset Curators

[More Information Needed]

Licensing Information

The dataset is available under the Creative Commons NonCommercial (CC BY-NC 4.0).

Citation Information

@misc{alpaca,
  author = {Rohan Taori and Ishaan Gulrajani and Tianyi Zhang and Yann Dubois and Xuechen Li and Carlos Guestrin and Percy Liang and Tatsunori B. Hashimoto },
  title = {Stanford Alpaca: An Instruction-following LLaMA model},
  year = {2023},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/tatsu-lab/stanford_alpaca}},
}

Contributions

[More Information Needed]