Card: add RunPod deploy section

Remove old-named safetensors

config: model.safetensors

inference: prefer model.safetensors

Rename projector to model.safetensors (param count)

Card: add composability demo section

Add composability demo figure

Remove pickle checkpoint (safetensors is primary)

config: weights_file -> safetensors

inference: load safetensors (pt back-compat)

Add projector as safetensors

Card: add badges

Card: add banner

Add k3-vision banner

results: full DINOv2 R@1/R@5/R@10

Fill DINOv2 comparison rows (no empty cells)

fusion-embedding-2-k3-vision v0.1-preview: project Kimi K3's MoonViT-V2 vision encoder into the fusion-embedding-2 space

initial commit