ramithuh commited on
Commit
bc21d11
1 Parent(s): a76dca4

Add `dataset_summary` and `dataset` sections

Browse files
Files changed (1) hide show
  1. README.md +35 -0
README.md CHANGED
@@ -182,4 +182,39 @@ size_categories:
182
  source_datasets:
183
  - original
184
  - extended
 
 
185
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
182
  source_datasets:
183
  - original
184
  - extended
185
+ task_categories:
186
+ - text-generation
187
  ---
188
+
189
+ ![Aya Header](https://huggingface.co/datasets/CohereForAI/aya_dataset/resolve/main/aya_header.png)
190
+
191
+ # Dataset Summary
192
+
193
+ `Aya Evaluation Suite` contains a total of 25,750 open-ended conversation-style prompts to evaluate multilingual open-ended generation quality.\
194
+ To strike a balance between language coverage and the quality that comes with human curation, we create an evaluation suite that includes:
195
+ 1) human-curated examples in 7 languages (`tur,eng,yor,arb,zho,por,tel`).
196
+ 2) machine-translations of handpicked examples into 101 languages.
197
+ 3) human-post-edited translations into 6 languages (`hin,srp,rus,fra,arb,spa`).
198
+
199
+ ---
200
+
201
+ - **Curated by:** Contributors of [Aya Open Science Intiative](https://aya.for.ai/), professional annotators, and synthetic generation
202
+ - **Language(s):** 101 languages
203
+ - **License:** [Apache 2.0](https://opensource.org/license/apache-2-0)
204
+ - **Aya Datasets Family:**
205
+ | Name | Explanation |
206
+ |------|--------------|
207
+ | [aya_dataset](https://huggingface.co/datasets/CohereForAI/aya_dataset) | Human-annotated multilingual instruction finetuning dataset, comprising over 204K instances across 65 languages. |
208
+ | [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 513k instances for various tasks.|
209
+ | [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.|
210
+
211
+
212
+ # Dataset
213
+
214
+ The `Aya Evaluation Suite` includes the following subsets:
215
+
216
+ 1. **aya-human-annotated**: 250 original human-written prompts in 7 languages each.
217
+ 2. **dolly-machine-translated**: 200 human-selected prompts from [databricks-dolly-15k](https://www.databricks.com/blog/2023/04/12/dolly-first-open-commercially-viable-instruction-tuned-llm)
218
+ , automatically translated with the [NLLB model](https://ai.meta.com/research/no-language-left-behind/) from English into 101 languages.
219
+ 3. **dolly-human-edited**: 200 dolly-machine-translated prompts post-edited by fluent speakers for 6 languages.
220
+