jphme/synthia_german_experimental

---
dataset_info:
features:
- name: system
dtype: string
- name: instruction
dtype: string
- name: response
dtype: string
- name: id
dtype: string
- name: score_deutsch
dtype: int64
- name: score_antwort
dtype: int64
- name: score_prompt
dtype: int64
splits:
- name: train
num_bytes: 275744686
num_examples: 105300
download_size: 133195884
dataset_size: 275744686
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---

This is a translated German version of the original [Synthia](https://huggingface.co/datasets/migtissera/Synthia-v1.3) dataset. It was translated with a model specifically created to translate instructions and "language" quality should be good, however there are known instances of faulty examples (e.g. wordgames that don't work for non-english languages), so please treat with care.
The scores are ratings of the instruction, answer and language quality on a scale from 1-3 and were created by a small experimental reward model. These are even more experimental (and the model was too small/weak to evaluate correctness), so use at your own risk.

If you use this dataset to train models, I would be thankful for an attribution (and please also attribute Migel Tissera for his great original dataset).

Improved/filtered versions of this dataset might be released by DiscoResearch in the future.

原始信息汇总

数据集概述

数据集特征

• system: 字符串类型
• instruction: 字符串类型
• response: 字符串类型
• id: 字符串类型
• score_deutsch: 整数类型（int64）
• score_antwort: 整数类型（int64）
• score_prompt: 整数类型（int64）

数据集分割

• train:
  • 数据量: 275744686 字节
  • 示例数量: 105300

数据集大小

• 下载大小: 133195884 字节
• 数据集大小: 275744686 字节

配置

• config_name: default
• data_files:
  • split: train
  • path: data/train-*