Datasets:

HuggingFaceTB
/

stack-edu

Modalities:

Formats:

Size:

ArXiv:

Libraries:

Dataset card Data Studio Files Files and versions Community

loubnabnl HF Staff commited on 26 days ago

Commit

118e0a6

·

verified ·

1 Parent(s): 8d2edf0

Update README.md

Files changed (1) hide show

README.md +1 -1

README.md CHANGED Viewed

@@ -497,7 +497,7 @@ To build Stack-Edu, we:
 - Trained 15 language-specific classifiers, using [StarEncoder](https://huggingface.co/bigcode/starencoder) model on synthetic annotations generated by Llama3-70B-Instruct. The classifiers for each language are available in this [collection](https://huggingface.co/collections/HuggingFaceTB/the-ultimate-collection-of-code-classifiers-67b5aa3eb8994a4b71453005).
 - Applied a filtering threshold of 3 (out of 5) to retain highly educational content, except for Java, which performed best at threshold 2.
-Stack-Edu, shows consistent improvement over StarCoder2data on all the programming languages on MultiPL-E benchmark.
 <img src="https://cdn-uploads.huggingface.co/production/uploads/61c141342aac764ce1654e43/GWnPgD0diMu0I8buK6mvG.png" width="600"/>

 - Trained 15 language-specific classifiers, using [StarEncoder](https://huggingface.co/bigcode/starencoder) model on synthetic annotations generated by Llama3-70B-Instruct. The classifiers for each language are available in this [collection](https://huggingface.co/collections/HuggingFaceTB/the-ultimate-collection-of-code-classifiers-67b5aa3eb8994a4b71453005).
 - Applied a filtering threshold of 3 (out of 5) to retain highly educational content, except for Java, which performed best at threshold 2.
+Stack-Edu shows consistent improvement over StarCoder2data on all the programming languages on MultiPL-E benchmark.
 <img src="https://cdn-uploads.huggingface.co/production/uploads/61c141342aac764ce1654e43/GWnPgD0diMu0I8buK6mvG.png" width="600"/>