-
Notifications
You must be signed in to change notification settings - Fork 274
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Questions about the free-law data used in the paper "Adapt LLM to domains" #164
Comments
Hi, The raw data size mentioned in Table 7 in our paper (51.2 GiB) is copied from Table 1 in the Pile paper: https://arxiv.org/pdf/2101.00027.pdf. I downloaded the data from Hugging Face: https://huggingface.co/datasets/EleutherAI/pile, which should be the same as yours. However, when looking at your link:
I noticed the term "partial" in the link. Does it mean you only downloaded a partial set of the entire dataset😂? I downloaded the data using the following Python code. Perhaps you can try this to download the full dataset: from datasets import load_dataset
free_law_data = load_dataset('EleutherAI/pile', 'free_law') |
Thank you very much for your reply. I'll try your method again |
Dear Authors, you have undoubtedly done an excellent job (domain-specific post-pre-training). But I have a small question about the size of the free-law data used in the original paper, I free downloaded from (https://huggingface.co/datasets/EleutherAI/pile/tree/refs%2Fconvert%2Fparquet/free_law/partial/train) - law data This seems to be much smaller than the 35G (16B tokens) described in the paper "Table 7", but only 1.4B tokens are actually processed using llama tokenizer. May I ask whether the author used the data in this link or another link?
The text was updated successfully, but these errors were encountered: