dkoplow/ZKL-CDHH
收藏Hugging Face2026-04-02 更新2026-04-12 收录
下载链接:
https://hf-mirror.com/datasets/dkoplow/ZKL-CDHH
下载链接
链接失效反馈官方服务:
资源简介:
---
pretty_name: ZKL CD+HH (parquet)
tags:
- rlhf
- countdown
---
# ZKL-CDHH
Parquet exports built from **Anthropic/hh-rlhf** (prompt-only) and
**Jiayi-Pan/Countdown-Tasks-3to4-Unique**. Rows are deduplicated; see the
generation script in the source repo (`gen_hf_ds.py`). You must comply with
the licenses and terms of the upstream datasets.
## Layout
- `val/data/val.parquet` — mixed validation split
- `train/<ratio>/data/train.parquet` — training mixes, e.g. `cd050-hh050`
## Columns
- `dataset`: `"hh"` or `"cd"`
- `prompt`: chat list of `{"role", "content"}` dicts
- `label`: Countdown ground truth list, or `null` for HH
- `id`: stable string id
## Load with `datasets`
```python
from datasets import load_dataset
repo = "dkoplow/ZKL-CDHH"
val = load_dataset(
"parquet",
data_files=f"hf://datasets/{repo}/val/data/val.parquet",
split="train",
)
train = load_dataset(
"parquet",
data_files=f"hf://datasets/{repo}/train/cd050-hh050/data/train.parquet",
split="train",
)
```
提供机构:
dkoplow



