language:
- de
task_categories: - conversational
- text-generation
dataset_info:
features:- name: raw
dtype: string - name: from
dtype: string - name: labels
dtype: string - name: conversations
list:- name: from
dtype: string - name: value
dtype: string
- name: from
- name: first_message
dtype: string - name: first_answer
dtype: string
splits: - name: train
num_bytes: 80567935.1091266
num_examples: 23275
download_size: 46600297
dataset_size: 80567935.1091266
configs:
- name: raw
- config_name: default
data_files:- split: train
path: data/train-*
- split: train
This dataset is an uncensored and massively cleaned, double checked merge of several german datasets / subsets
The mission of this work is building an high quality dataset for the german llm community.
This repo is continously updated and old parts being replaced with never.
Quality for Quantity
https://github.com/flozi00/chat-data-experiments/blob/main/chat_combiner.py