revert: restore server/app.py to last known working state

Deploy HTML landing page at root / (replaces JSON response)

Update README.md

Add direct links to training script, log CSV, and plots in README

revert: restore requirements.txt to last working state

revert: restore server/app.py to last working state

fix: remove openenv.core dependency — pure FastAPI, graceful fallback

fix: remove openenv.core dependency — pure FastAPI, graceful fallback

fix: add openenv>=0.1.13 to requirements.txt

fix: redirect root / to /ui so judges see Gradio UI not raw JSON

link: add real blog post URL to README

add: complete HF blog post content for hackathon submission

story: add qualitative reward examples, blog+video links, before/after proof

results: update with real v3 eval data — ECE -86.5% on GPQA-Lite, reward 0.15→0.75

update README: real measured eval — ECE 0.069→0.048, 100Q, 7 domains

update README: real results table + embed training_curves and baseline_vs_trained plots

update README with live Space URL, adapter link, real training status

feat: A10G-optimised GRPO config — 256 tokens, bf16, 300 samples

feat: sanity check cell, format-enforcing system prompt, auto HF Hub push after training

data: add 7534 real tasks cache (GSM8K, TriviaQA, ARC, SciQ, MedMCQA)

fix: remove unsloth/trl/bitsandbytes from Space requirements (training only, use Colab)

fix: pure FastAPI on port 7860 — all OpenEnv endpoints live + Gradio at /ui

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Upload folder using huggingface_hub

initial commit