Relation extraction is the fundamental task in the field of natural language processing, which aims to identify the relations between entities in a sentence. Fine-grained relation extraction is an important extension of this task, aiming to identify the fine-grained relations between entities in a sentence. Existing knowledge-enhanced fine-grained relation extraction methods primarily link entities in the sentence to the knowledge graph through entity linking, thereby obtaining representations of the entities with more semantic information. Unfortunately, both of the prerequisites including semantically rich knowledge graphs and high-quality entity linkers are hard to satisfy in practical circumstances. Recently, the advent of large language models (LLMs) like GPT-4 and LLaMA have opened up new chances for knowledge generation. However, the challenge lies in fully harnessing the potential of these large language models to generate high-quality knowledge. To address these challenges, we propose a multi-agent collaborative knowledge generation framework for entity knowledge generation to provide entity knowledge for knowledge-enhanced sentence-level fine-grained relation extraction models. We conduct several experiments on comparing existing knowledge-enhanced and non-knowledge-enhanced baselines with our framework and ablation study to demonstrate the effectiveness of our proposed method.