Upload code.png

Restore KV-cached decode (O(n) not O(n^2)) with exact engram window; stream output; load model at startup

space card v2

update to Byrne-Docling v2 (letterbox + atomic DocTags tokens); pulls new weights from model repo

Upload app.py with huggingface_hub

Upload generate.py with huggingface_hub

Upload app.py with huggingface_hub

Upload app.py with huggingface_hub

Upload code.png

Delete examples/code.png

space card

Byrne-Docling demo (code only; pulls weights from model repo)

initial commit