About the dataset

[sunxin000]

Why is the default value for the dataset set to Ori/llama-2-13b-peft-nq-no-ret, a Hugging Face model?

parser.add_argument(
        "--dataset_name",
        type=str,
        default="Ori/llama-2-13b-peft-nq-no-ret",
    )

[oriyor]

hi! this looks like a copying error, please take a look at the available datasets at the HuggingFace collection - https://huggingface.co/collections/Ori/retrobust-65198eef2b4fffcb4100e163

[sunxin000]

Thank you for your kind response. However, I am still a bit confused. Why doesn't the training data contain information about retrieval?