metadata
dataset_info:
features:
- name: conversations
list:
- name: from
dtype: string
- name: value
dtype: string
splits:
- name: train
num_bytes: 19887468
num_examples: 7020
download_size: 9880827
dataset_size: 19887468
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
This dataset by AbacusAI was crafted by Eric Hartford This is a synthetic dataset, generated mainly with Mistral-Medium and dolphin-2.7-mixtral-8x7b The purpose of this dataset is to train the model to respect the System Prompt throughout the entire conversation, no matter how unconventional the system prompt might be. This dataset is under continued development - my intent is to grow it to 100k conversations. But, for now, it is good enough to start using.
AbacusAI 的这个数据集由 Eric Hartford 制作 这是一个合成数据集，主要由 Mistral-Medium 和 dolphin-2.7-mixtral-8x7b 生成。 该数据集的目的是训练模型在整个对话过程中都尊重系统提示，无论系统提示多么不合常规。 这个数据集还在继续开发中--我的目标是将其增加到 10 万个对话。 但现在，它已经足够好，可以开始使用了。