Characters144
Words17
~Tokens36
Size144 B
Conventional ViTs assume fixed input geometry, which introduces inefficiencies and distortions when scaling to real-world multimodal workloads.

nostalgicgareth
Open-MoonViT is a simple, single-file PyTorch implementation of the Vision Transformer from the Kimi-VL paper.
Conventional ViTs assume fixed input geometry, which introduces inefficiencies and distortions when scaling to real-world multimodal workloads.
Loading chart...
Scroll to load comments...
Loading recommendations...