Datasets:
Inquiry Downloading the PDF Dataset
I hope this message finds you well.
I recently noticed that your team released a PDF dataset a few days ago. Given the scarcity of open-source PDF data, this release is incredibly valuable—especially for model training—and I’m truly excited about the opportunity to work with it.
However, I’ve encountered some difficulties while trying to download the dataset. The download method described in your documentation doesn’t seem to be working (loading a specific split of the dataset), and when I attempted to download the file directly from the website and unzip it, I found that the archive appears to be corrupted.
Would you kindly be able to double-check the dataset files or let me know if I might have made an error during the download process? Any guidance you could provide would be greatly appreciated.
Thank you very much for your time and support!
Hi there - thanks for the feedback and sorry for the trouble. We are actively working on this issue and will get back to you ASAP with a fix!
@KikiQi I was able to reproduce the error. Thank you so much for pointing this out -- we recently changed the name/structure of our repository so the code provided in the README was outdated. I have now updated it, and this should now work, at least to load it in:
dataset = load_dataset("allenai/dolma3_pool",
data_files="data/olmocr_science_pdfs-*/*.jsonl.zst",
split="train")
The second error here is what I'm working on now - it looks like a few of the files were prematurely truncated during processing. I'm working on finding those files and fixing them - will update you soon. Thanks so much for your patience.