Comment by siddheshgunjal

At my work, we still prefer to use distilbert for text classification. It almost always does well with a little bit of fine tuning. In very rare cases, we use LLMs/Agentic setup when the task involves refering both images and text and the same time.