Is One-Shot In-Context Learning Helpful for Data Selection in Task-Specific Fine-Tuning of Multimodal LLMs?

Xiao An, Jiaxing Sun, Ting Hu +1

Injecting world knowledge into pretrained multimodal large language models (MLLMs) is essential for domain-specific applications. Task-specific fine-tuning achieves this by tailori…