Update README.md

README.md

@@ -158,6 +158,7 @@ This dataset can be used to train, finetune, and evaluate multilingual LLMs.
   |------|--------------|
   | [aya_dataset](https://huggingface.co/datasets/CohereForAI/aya_dataset) | Human-annotated multilingual instruction finetuning dataset, comprising over 204K instances across 65 languages. |
   | [aya_collection](https://huggingface.co/datasets/CohereForAI/aya_collection) | Created by applying instruction-style templates from fluent speakers to 44 datasets, including translations of 19 instruction-style datasets into 101 languages, providing 513M instances for various tasks.|
+  | [aya_collection_language_split](https://huggingface.co/datasets/CohereForAI/aya_collection_language_split) | Aya Collection structured based on language level subsets. |
   | [aya_evaluation_suite](https://huggingface.co/datasets/CohereForAI/aya_evaluation_suite) | A diverse evaluation set for multilingual open-ended generation, featuring 250 culturally grounded prompts in 7 languages, 200 translated prompts in 24 languages, and human-edited versions selected for cross-cultural relevance from English Dolly in 6 languages.|
   | [aya_redteaming](https://huggingface.co/datasets/CohereForAI/aya_redteaming)|  A red-teaming dataset consisting of harmful prompts in 8 languages across 9 different categories of harm with explicit labels for "global" and "local" harm.|