This file contains the embeddings for the full fineweb-edu dataset.

The dataset has been deduplicated (using only exact deduplication).

The emb folder contains for each parquet file a new_{parquet_name}.npy and old_{parquet}.npy file. The old refers to text that has been seen in the smaller 10B/100B/350B data samples.

Cohere embed-multilingual-v3.0 model has been used.

The index of the dataset can be found here:
https://huggingface.co/datasets/Cohere/fineweb-edu-index

The corpus can be found here:
https://huggingface.co/datasets/Cohere/fineweb-edu-corpus