Simonkami/bvb_selftrain3
收藏Hugging Face2026-04-28 更新2026-05-03 收录
下载链接：
https://hf-mirror.com/datasets/Simonkami/bvb_selftrain3
下载链接
链接失效反馈官方服务：
资源简介：
---
configs:
- config_name: default
data_files:
- split: train
path: train.jsonl
---
# BvsALL_explained_selftrain_round1_half_original
This dataset is a local mixed SFT dataset built from:
- Original dataset: `Simonkami/bvb_8000` split `train`
- Self-training file: `selftrain_good_samples_round1.jsonl`
## Mixing settings
- Original total: 8840
- Original used: 2652
- Original fraction: 0.3
- Original max samples: 0
- Selftrain raw total: 49162
- Selftrain valid: 49162
- Selftrain after action balance: 6592
- Selftrain used: 6187
- Target selftrain ratio: 0.7
- Actual selftrain ratio: 0.7000
- Max selftrain samples per action: 400
- Final total: 8839
## Columns
- `instruction`
- `input`
- `output`
`train_with_meta.jsonl` contains extra `source` and `score` fields for inspection, but `train.jsonl` is the recommended file for SFT training.
提供机构：
Simonkami
