Suggestion: Update benchmark table with Canarim dataset
Hello! Thank you for sharing the Aurora-PT dataset.
I noticed that the README mentions Aurora-PT is currently the largest dataset available for Portuguese. However, there is another public repository called Canarim (dominguesm/canarim), published in early 2024, which contains over 342 million documents and totals around 1.09 TB (uncompressed/processed text).
It might be a great addition to include it in your benchmark table to make the comparison even more comprehensive for the community!
- Dataset: dominguesm/canarim
- Size: ~1.09 TB
- Documents: ~342.8 Million
Best regards!
Hello!
On the same note, please consider updating the information from GigaVerbo to GigaVerbo-v2!
We have done a lot to improve from our first version, now with 320 billion tokens (counted with a proper Portuguese tokenizer) and ~375 million documents (after global deduplication). All with educational quality and toxicity annotations. 🤗
Thank you for your suggestions!
We are reviewing them to determine how we can move forward with the Aurora Card.
No update on this? 🤔
Thank you for your suggestions.
We reviewed them internally and decided to include references to both datasets in the dataset card.
If you have any further questions or comments, please feel free to reach out to us here.
Thank you! And congratulations for the work 🤗