4 ms·
I got excited by reading the article about releasing the training data, went to their HF account to look at the data (dolma3) and first rows? Text scraped from
by Oras 11mo ago
I got excited by reading the article about releasing the training data, went to their HF account to look at the data (dolma3) and first rows? Text scraped from porn websites!
https://huggingface.co/datasets/allenai/dolma3 https://huggingface.co/datasets/allenai/dolma3
- logicchains 11mo agoErotic fiction is one of the main use cases of such models.
- andy99 11mo agoIsn’t this before any curation has happened? I looked at it, I can see why it looks bad, but if they’re really being open about the whole pipeline, they have to include everything. Giving them a hard time for it only promotes keeping models closed. That said I like to think of it was my dataset I would have shuffled that part down in the list so it didn’t show up on the hf preview
- Oras 11mo agoHard time? What value does adult videos description, views and comments add to small (7,32B) models?
- andy99 11mo agoIt says it’s common crawl, I interpret it to mean this is a generic web scrape dataset, presumably they filter stuff out they don’t want before pretraining. You’d have to do do some ablation testing to know what value it adds
- ccgreg 11mo agoCommon Crawl is a particular dataset. commoncrawl.org
- khimaros 11mo agowhat if that's where they learned how to utilize the double entendre? hard times indeed.