New York Site views —Page views —
← All series

Vision-language models

Follow images through encoders and projectors, then compare multi-image and high-resolution inputs.

Reading order

  1. How an image enters a language model: encoders, projectors and alignment
  2. How LLaVA, Qwen-VL, MiniCPM-V and InternVL differ
  3. Working with multiple images, high resolution and long illustrated documents in VLMs