lightblue/rag_multilingual_training_negatives
How this dataset was made We trained on chunks sourced from the documents in MADLAD-400 dataset that had been evaluated to contain a higher amount of educational information according to a state-of-the-art LLM. We took chunks of size 250 tokens, 500 tokens, and 1000 tokens randomly for each document. We then used these chunks to generate questions and answers based on this text using a state-of-the-art LLM. Finally, we selected negatives for each chunk using the similarity from… See the full description on the dataset page: https://huggingface.co/datasets/lightblue/rag_multilingual_training_ne
| Author / organization | lightblue |
|---|---|
| Downloads | 845 (as reported by Hugging Face) |
| Likes | 3 |
| Created | 3 September 2024 |
| Last updated at source | 23 January 2025 |
Popularity, last 3 days
845 now (+38 since 2026-10-16). Counts as reported by the source, snapshotted daily by Silytics.
Recent changes
- 17 Oct 2026 Description: How this dataset was made We trained on chunks sourced from the documents in MADLAD-400 dataset that had been evaluated to contain a higher amount of educational information according to a state-of-the-art LLM. We took chunks of size 250 to → How this dataset was made We trained on chunks sourced from the documents in MADLAD-400 dataset that had been evaluated to contain a higher amount of educational information according to a state-of-the-art LLM. We took chunks of size 250 to
Changes detected by Silytics between refreshes of the source.
More from lightblue
Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 18 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.