Fine-tuning vision-language models (VLMs) like CLIP to downstream tasks is often
necessary to optimize their performance. However, a major obstacle is the limited availability …