Datasets:

HuggingFaceFW
/

fineweb-edu

Tasks:

Text Generation

Modalities:

Formats:

Languages:

Size:

ArXiv:

DOI:

Libraries:

License:

Dataset card Viewer Files Files and versions Community

anton-l HF staff commited on May 31, 2024

Commit

20b2c9f

·

verified ·

1 Parent(s): 6861e17

Update README.md

Files changed (1) hide show

README.md +2 -2

README.md CHANGED Viewed

@@ -415,7 +415,7 @@ configs:
 ## What is it?
-📚 FineWeb-Edu  dataset consists of **1.3T tokens**  and  **5.4T tokens** ([FineWeb-Edu -Large](https://huggingface.co/datasets/HuggingFaceFW/fineweb-edu-large)) of educational web pages filtered from 🍷 FineWeb dataset. This is the 1.3 trillion version.
 To enhance FineWeb's quality, we developed an [educational quality classifier](https://huggingface.co/HuggingFaceFW/fineweb-edu-classifier) using annotations generated by LLama3-70B-Instruct. We then used this classifier to retain only the most educational web pages. FineWeb-Edu outperforms FineWeb on popular benchmarks and shows the power of classifiers trained on synthetic data.
@@ -499,7 +499,7 @@ The classifier is available at: [https://huggingface.co/HuggingFaceFW/fineweb-ed
 We filtered out samples with scores lower than 3. This removed 92% of the dataset, leaving us with 1.2T educational tokens. Our ablation demonstrated that this refined dataset significantly outperforms the original FineWeb dumps and even the best dump, FineWeb-2024-10. To retain more tokens, we also experimented with a less strict threshold of 2 instead of 3. This approach preserved 4.5T tokens and still outperformed the non-filtered dataset.
 TODO: add ablation results
-We release these two dataset as [FineWeb-Edu](https://huggingface.co/datasets/HuggingFaceFW/fineweb-edu) and [FineWeb-Edu-Large](https://huggingface.co/datasets/HuggingFaceFW/fineweb-edu-large) along with the classifier.
 ## Dataset performance evaluation and ablations

 ## What is it?
+📚 FineWeb-Edu  dataset consists of **1.3T tokens**  and  **5.4T tokens** ([FineWeb-Edu-score-2](https://huggingface.co/datasets/HuggingFaceFW/fineweb-edu-score-2)) of educational web pages filtered from 🍷 FineWeb dataset. This is the 1.3 trillion version.
 To enhance FineWeb's quality, we developed an [educational quality classifier](https://huggingface.co/HuggingFaceFW/fineweb-edu-classifier) using annotations generated by LLama3-70B-Instruct. We then used this classifier to retain only the most educational web pages. FineWeb-Edu outperforms FineWeb on popular benchmarks and shows the power of classifiers trained on synthetic data.
 We filtered out samples with scores lower than 3. This removed 92% of the dataset, leaving us with 1.2T educational tokens. Our ablation demonstrated that this refined dataset significantly outperforms the original FineWeb dumps and even the best dump, FineWeb-2024-10. To retain more tokens, we also experimented with a less strict threshold of 2 instead of 3. This approach preserved 4.5T tokens and still outperformed the non-filtered dataset.
 TODO: add ablation results
+We release these two dataset as [FineWeb-Edu](https://huggingface.co/datasets/HuggingFaceFW/fineweb-edu) and [FineWeb-Edu-score-2](https://huggingface.co/datasets/HuggingFaceFW/fineweb-edu-score-2) along with the classifier.
 ## Dataset performance evaluation and ablations