This dataset is based on the Japanese version of Wikipedia dataset and converted into a multi-turn conversation format using llama2Pro8B.
Since it is a llama2 license, it can be used commercially for services.
Some strange dialogue may be included as it has not been screened by humans.
We generated over 80,000 conversations 22 days on an A100 80GBx7 machine and automatically screened them.
