Rename model display: kpaa-privacy-ko -> 개인정보 미니 상담 AI

Serve split HTML directly at / (no redirect) — overrides Gradio's /

Redirect / -> /split (insert route at index 0 to override Gradio's /)

Add /split route — Open WebUI iframe + 참고자료 polling layout

Use generator pattern (cross-process yield via spaces res_queue) — ZeroGPU forks process so streamer queue isn't shared

Force streamer.end() in finally — transformers 5.x doesn't always auto-call it

Hybrid: demo.launch() (ZeroGPU works) + post-launch attach KPAA /v1 routes to demo.app

MINIMAL TEST: Gradio + @spaces.GPU + demo.launch (HF ZeroGPU canonical pattern)

Fix diag import path (spaces.config not spaces.zero.config); SPACES_ZERO_GPU=1 confirmed

Diagnostic: print SPACES_ZERO_GPU env + canary marking status

Use mount_gradio_app(ssr_mode=False) — explicit param overrides HF env var to skip Node port bind

Manual Gradio sub-app mount (skip startup hook that binds 7861)

Re-add Gradio Blocks + mount, with GRADIO_SSR_MODE=False to avoid port conflict

Drop Gradio mount entirely; pure FastAPI on uvicorn (avoids SSR port conflict)

Hide demo inside main() to avoid HF auto-launch port conflict

Add backend README with frontmatter (zero-a10g)

Add backend-specific app.py (FastAPI + Gradio status)

Initial backend code: src/kpaa, runtime data, requirements

initial commit