CATIE is pleased to present 𝐃𝐅𝐏 (𝐃𝐚𝐭𝐚𝐬𝐞𝐭 𝐨𝐟 𝐅𝐫𝐞𝐧𝐜𝐡 𝐏𝐫𝐨𝐦𝐩𝐭𝐬).
This dataset of French prompts contains more than 𝟏𝟎𝟎𝐌 𝐥𝐢𝐧𝐞𝐬 to train an LLM-type model.
It covers more than 𝟑𝟎 𝐝𝐢𝐟𝐟𝐞𝐫𝐞𝐧𝐭 𝐍𝐋𝐏 𝐭𝐚𝐬𝐤𝐬.
The prompts handle imperative forms (commands) as well as the informal “tu” and formal “vous”, to cover as much as possible of the pre-training data used by the model that will use DFP.
You can find this dataset here
The dataset card contains everything you need to know: the list of tasks covered, the prompts used, and the licenses and citations.
As an example of how this dataset can be used, 𝐰𝐞 𝐟𝐢𝐧𝐞-𝐭𝐮𝐧𝐞𝐝 𝐚 𝐌𝐢𝐬𝐭𝐫𝐚𝐥-𝟕𝐁-𝐯𝟎.𝟏 𝐦𝐨𝐝𝐞𝐥 on it.
It is also available here