Incorrect dataset license for Pythia-2.8B

[aphedges]

The Pythia-2.8B page states, "Datasets [Apache-2.0]", but the linked dataset page has no mention of the Apache license, and the paper linked on the dataset page describes data sources that are not open. The problem likely applies to the other Pythia models as well.

I came across #168, which would likely better resolve the problem, but in the meantime, I feel it's misleading to leave the non-open dataset documented as being under an open license.

[aphedges]

I've noticed a similar problem with OLMo-1B-hf. The linked paper, "OLMo: Accelerating the Science of Language Models", says the dataset includes Common Crawl and Reddit, neither of which are open.

[lehors]

Thanks for pointing this out! We'll look into it and update our data.
Much appreciated.