fix: bound Space OCR runtime

fix: avoid truncating CPU OCR output

fix: patch causal mask signature

fix: patch Gradio API info schema

fix: keep legacy OCR API alias

fix: use non-queued OCR event

fix: lazy load model in Space app

fix: stabilize CPU Space OCR demo

docs: remove dangling reference to private eval-set link

fix(space): add einops (hard dep of custom modeling code)

fix(space): add sentencepiece for LLaMA tokenizer

fix(space): pin transformers==4.55.0 + use AutoModelForCausalLM

feat(demo): port eval post-processor, CPU feedback, doc updates

fix: add torchvision (AutoImageProcessor backend requirement)

upload .gitignore

upload README.md

upload requirements.txt

upload app.py

initial commit