Spaces:
Runtime error
Runtime error
Commit History
-
fix graders
UjjwalPardeshif1b57dc
-
minor changes
omkarrr88206438f -
Minor fixes
omkarrr887336adb
-
Major fixes
omkarrr886be6d8e
-
task 6 and 7 made hard
omkarrr8802e58fe -
fix: remove parametric fallback from simulation.py, clean baseline table
omkarrr88- simulation.py now uses ONLY real PyTorch training (run_real_training()) Zero torch.exp(), zero torch.sigmoid(), zero synthetic curves. All loss/accuracy curves come from actual forward+backward passes. - Removed test_simulation_fallback.py (tested deleted parametric code) - README: removed incomplete Llama 70B column, fixed task_007 to "Hard" - 239 tests passing, openenv validate passing05ccdc6
-
feat: harden Task 5 grader + fix inference.py for hackathon submission
omkarrr88- Task 5 grader now requires weight+data inspection for full credit (1.0) Heuristic drops from 0.96 to 0.80. Red herring penalty (-0.20) + wrong-fix penalty (-0.10) mean frontier LLMs that chase false leads score 0.38-0.48. - inference.py: add 30s API timeout, log grader errors instead of swallowing, add WebSocket ping/timeout for reliability - deploy-hf.sh: exclude .env, .coverage, uv.lock from rsync to prevent leaking API keys to public HF Space - requirements.txt: add openai and websockets as explicit deps - Updated baseline scores in README to reflect new grader - 258 tests passing, openenv validate passing56b42b0
-
Real training curves added
omkarrr88aa0bed2 -
Task 7 added
omkarrr880b9b77b -
Remaining task added + full openenv compliance
omkarrr889e6a926 -
Version 1
omkarrr88e2f8b29