Dataset Card for uplimit-synthetic-data-week-2-with-multi-turn

This dataset has been created with distilabel.

The pipeline script was uploaded to easily reproduce the dataset: colab_kernel_launcher.py.

It can be run directly using the CLI:

distilabel pipeline run --script "https://huggingface.co/datasets/Tiamz/uplimit-synthetic-data-week-2-with-multi-turn/raw/main/colab_kernel_launcher.py"

Dataset Summary

This dataset contains a pipeline.yaml which can be used to reproduce the pipeline that generated it in distilabel using the distilabel CLI:

distilabel pipeline run --config "https://huggingface.co/datasets/Tiamz/uplimit-synthetic-data-week-2-with-multi-turn/raw/main/pipeline.yaml"

or explore the configuration:

distilabel pipeline info --config "https://huggingface.co/datasets/Tiamz/uplimit-synthetic-data-week-2-with-multi-turn/raw/main/pipeline.yaml"

Dataset structure

The examples have the following structure per configuration:

Configuration: default

---

{
    "chosen": null,
    "conversation": [],
    "distilabel_metadata": {
        "raw_input_chat_generation_0": null,
        "raw_input_chat_generation_1": null,
        "raw_output_chat_generation_0": null,
        "raw_output_chat_generation_1": null,
        "statistics_magpie_generator_0": {
            "input_tokens": [
                85,
                85,
                85
            ],
            "output_tokens": [
                0,
                0,
                0
            ]
        }
    },
    "model_name": null,
    "rejected": null
}

This subset can be loaded as:

from datasets import load_dataset

ds = load_dataset("Tiamz/uplimit-synthetic-data-week-2-with-multi-turn", "default")

Or simply as it follows, since there's only one configuration and is named default:

from datasets import load_dataset

ds = load_dataset("Tiamz/uplimit-synthetic-data-week-2-with-multi-turn")

References

@misc{xu2024magpiealignmentdatasynthesis,
    title={Magpie: Alignment Data Synthesis from Scratch by Prompting Aligned LLMs with Nothing},
    author={Zhangchen Xu and Fengqing Jiang and Luyao Niu and Yuntian Deng and Radha Poovendran and Yejin Choi and Bill Yuchen Lin},
    year={2024},
    eprint={2406.08464},
    archivePrefix={arXiv},
    primaryClass={cs.CL},
    url={https://arxiv.org/abs/2406.08464},
}