MemGPT/MemGPT-DPO-Dataset

MemGPT-DPO-Dataset是一个用于文本生成模型微调的数据集，特别是用于DPO（直接偏好优化）和SFT（监督微调）微调。该数据集由GPT-4生成，包含42,293行数据，仅有一个训练集分割。数据集的目标是教导大型语言模型（LLM）在MemGPT特定工具中选择正确的函数。数据集的生成过程包括快速手动检查和清理，以确保数据质量。数据集的使用场景主要是为了提升开源模型在MemGPT上的性能，使其在函数调用方面超越GPT-4。

MemGPT-DPO-Dataset is a dataset designed for fine-tuning text generation models, specifically for DPO (Direct Preference Optimization) and SFT (Supervised Fine-Tuning) fine-tuning tasks. It was generated by GPT-4, containing 42,293 rows of data with only a single training split. The core objective of this dataset is to teach large language models (LLMs) to select the correct functions when using MemGPT-specific tools. Rapid manual inspections and cleaning were conducted during the dataset generation process to ensure data quality. The primary application scenario of this dataset is to enhance the performance of open-source models on MemGPT, enabling them to outperform GPT-4 in terms of function calling capabilities.

原始信息汇总

MemGPT-DPO-Dataset 数据集概述

数据集名称

• 名称: MemGPT-DPO-Dataset

数据集版本

• 版本: 初始版本

数据集系列

• 系列: 可能的系列数据集中的首个发布