internlm/Agent-FLAN
收藏资源简介:
Agent-FLAN数据集旨在通过微调大型语言模型(LLMs)来提升其在代理任务中的表现。该数据集由AgentInstruct、Toolbench和定制的负面代理样本组成,采用特定的数据生成管道进行训练。数据集的特征包括对话格式,每个对话包含角色、内容和损失等字段。通过精心设计的训练语料库,Agent-FLAN使Llama2-7B模型在各种代理评估数据集上表现优于之前的最佳工作,并显著减轻了幻觉问题。
The Agent-FLAN dataset is developed to improve the performance of large language models (LLMs) on agentic tasks through fine-tuning. It consists of AgentInstruct, Toolbench, and custom negative agent samples, and is trained via a dedicated data generation pipeline. The dataset features a dialogue format, where each dialogue contains fields including role, content, and loss. With its carefully curated training corpus, Agent-FLAN enables the Llama2-7B model to outperform prior state-of-the-art approaches across diverse agent evaluation benchmarks, and substantially alleviates the hallucination issue.
数据集概述
数据集名称
- AgentFLAN
数据集配置
- 默认配置 (
default):- 包含多个数据文件,每个文件对应不同的数据分割:
agent_instruct_reactagent_instruct_tflantoolbench_instruct_j1s1_3ktoolbench_negativetoolbench_react_10ptoolbench_tflan_60p_r10r5u7toolbench_tflan_cot_30p
- 包含多个数据文件,每个文件对应不同的数据分割:
数据集信息
- 特征 (
features):- conversation:
- role (字符串类型)
- content (字符串类型)
- loss (布尔类型)
- id (字符串类型)
- conversation:
许可证
- Apache-2.0
标签
- agent




