yatin-superintelligence/digital-hospital-environment

Refine Digital Hospital Space frontend

Polish Space landing page

Refresh Digital Hospital HF Space

feat: session isolation + hf-space API backend + rollout mode

feat: allow ENV_URL/INFERENCE_URL env var overrides in collect_rollouts.sh

Upload server/app.py with huggingface_hub

chore: cleanup environment references and reorganize baseline trajectories

Remove TRAINING.md from Space — training files stay local only

Sync Digital Hospital Train: env files, updated gitignore, remove training files

Add TRAINING_TPU.md: full TPU-only GRPO strategy (EasyDeL + Kaggle v5e-8 + gemma-4-E2B-it)

TRAINING.md: add log_prob_old, correct GRPO loss formula, clarify dense reward advantage

Update TRAINING.md: offline rollout (Kaggle) + A100 gradient split, 4 parallel Spaces

Add TRAINING.md — GRPO Option B strategy document

Fix: accurate per-role phase descriptions + baseline card moved below treatment grading

Fix: show all 11 role baseline scores on landing page (was showing 5)

V10: Production-ready baseline — Digital Hospital Environment