Spaces:
Running
Running
Commit History
-
Fix Phase 2 validation: add Rubric graders + correct log format Running
G Rohith Yadav +1: Claude Sonnet 4.6Two root causes for the submission failure: 1. inference.py emitted JSON objects ([START] {...}) instead of the required key=value format ([START] task=X env=Y model=Z). The Phase 2 validator parses [START] task= lines to enumerate tasks — none were found, hence "Not enough tasks with graders". Fixed: each task now emits its own [START]/[STEP]s/[END] block using the exact key=value format the validator expects. 2. _compute_score() returned 0.0 (before traffic flows) and was capped at 0.999 (which formats as "1.00" with {:.2f}), both out of the required strictly-(0,1) range. Fixed: floor changed to 0.01, cap changed to 0.99. 3. SreDefenderEnvEnvironment never set self.rubric — the Environment base class exposes `rubric` for per-task grader introspection. Added SreDefenderRubric with 3 named child rubrics (block_single_ip, rate_limiter, payload_defender), one per hackathon task. step() now calls _apply_rubric() per OpenEnv spec. Co-Authored-By: C…a230d7c -
Fix: Laplace Smoothing to keep scores between 0 and 1
G Rohith Yadav0d370bc -
Fix: Multi-mode deployment for validate
G Rohith Yadavb2e586d -
SRE Defender Env - Meta x Scaler Hackathon
G Rohith Yadav68d3e5b