documentation/docs/tooltips/experiments/_limit_chained_samples.mdx

If set to True, model will be only trained on the full chained sample in case of nested conversations. If set to False, each separate sample is used for training.