Abstract: Prompt tuning has proven to be an effective alternative for fine-tuning the pre-trained vision-language models (VLMs) to downstream tasks. Among existing approaches, class-shared prompts ...