Datasets:

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

which language contains in this dataset

#16
by lucasxx - opened

which languages contains in this dataset

Ai2 org

Dolma contains English and code, though in practice you'll find some other languages because language detectors and filters are not perfect.

It is very bad news. I thought that the dataset used by OLMO project is democratizing the LLM related research. The first step of this would be involving low resource languages as well.

Ai2 org

We have to make sure we use our resources in the most effective way. Nobody benefits from a model that has a lot of features, but is mediocre at all of them. We talked about making multilingualism a headline feature of OLMo, but found that we didn't have enough interest and expertise on the team. If we had done it, it would have been an afterthought.

If you want to train a multilingual OLMo, we'd be happy to support you!

amanrangapur changed discussion status to closed

Sign up or log in to comment