Datasets:

Languages:
English
Size Categories:
n>1T
ArXiv:
License:

Question: How can I search the Dolma dataset without downloading it?

#21
by alishasrivas - opened

Hello,

I have a bunch of text files of books from Project Guttenberg and outside of it, I'd like to search the Dolma dataset to see if they are in Dolma. Do you provide a tool to do this effectively?

The only similar method I found for this is the _generate_examples(self, files: List[str]) in Dolma.py but in my understanding, it samples from a downloaded file which isn't what I'm looking for.

Thank you!

Hello!

You might be interested in the infini-gram search engine. It supports very fast n-gram counting and document searching in massive text corpora.

We currently support popular corpora like RedPajama (1.4T tokens), Pile (380B tokens), and C4 (200B tokens). We also support the official Dolma v1.6 sample (8B tokens). We're working on indexing the full Dolma and it should be available within the next month.

Best,
Jiacheng

Allen Institute for AI org

We have an internal search engine. I'll see if we can make that public, but no promises. You can get some statistics here? https://wimbd.apps.allenai.org

Sign up or log in to comment