Simonkami/bvb_selftrain3
收藏资源简介：
该数据集是一个本地混合的监督微调（SFT）数据集，由两部分数据混合而成：原始数据集`Simonkami/bvb_8000`的`train`分割和自训练文件`selftrain_good_samples_round1.jsonl`。混合设置中，原始数据使用了2652条（占总量的30%），自训练数据使用了6187条（目标比例为70%）。数据集包含`instruction`、`input`和`output`三列，并提供了一个包含额外字段`source`和`score`的文件用于检查，但推荐使用`train.jsonl`进行SFT训练。
This dataset is a local mixed SFT dataset built from the original dataset `Simonkami/bvb_8000` split `train` and the self-training file `selftrain_good_samples_round1.jsonl`. In the mixing settings, 2652 original samples (30% of the total) and 6187 self-training samples (target ratio 70%) were used. The dataset includes columns `instruction`, `input`, and `output`, and provides a file with additional fields `source` and `score` for inspection, but `train.jsonl` is recommended for SFT training.
