| --- |
| license: apache-2.0 |
| language: |
| - en |
| task_categories: |
| - text-generation |
| tags: |
| - synthetic |
| - datainventor |
| size_categories: |
| - n<1K |
| configs: |
| - config_name: default |
| data_files: |
| - split: train |
| path: data/train.jsonl |
| pretty_name: "DataInventor: African questions and answers" |
| --- |
| |
| # DataInventor: African questions and answers |
|
|
| A synthetic instruction dataset of 195 rows in English, built with [DataInventor](https://huggingface.co/spaces/LocalLLaMA/data-inventor), an open-source recreation of the Invent a Dataset workflow. Describe the data you want and DataInventor writes it. This is one of eight showcase datasets. |
|
|
| ## Request |
|
|
| > Dataset containing question-and-answer pairs covering diverse topics related to African research, literature, history, and current events. |
|
|
| The wording is one of the eight example requests in the Invent a Dataset technical report from Adaption Labs. This dataset is not affiliated with Adaption Labs and does not contain any of their output. |
|
|
| ## How it was made |
|
|
| 1. A blueprint was drafted from the request: topics, who asks, who answers, and the shapes a question can take. |
| 2. Unique combinations of topic, persona, shape, difficulty and length were sampled. |
| 3. For each combination, one message was written, then the reply. |
| 4. Rows were rejected for cut-off or degenerate text, for asking for many items at once, and for near-duplicate questions. |
|
|
| Generated with optillm. |
|
|
| ## Columns |
|
|
| `prompt`, `completion`, plus `area`, `subtopic`, `topic`, `persona`, `shape`, `difficulty` and `language` describing how each row was planned. |
|
|
| ## Quick numbers |
|
|
| - Rows: 195 |
| - Distinct 3-grams in prompts: 0.86 |
| - Average words per prompt: 50.7 |
| - Average words per completion: 349.6 |
|
|
| ## Limits |
|
|
| - Every row was written by a language model and has not been checked by a person. Replies can contain errors and invented facts. Check them before you train on this data or rely on it. |
| - Synthetic data repeats the habits and blind spots of the model that wrote it. |
|
|
| Make your own with [DataInventor](https://huggingface.co/spaces/LocalLLaMA/data-inventor). License: Apache 2.0. |
|
|
