Commit History

  1. remove ambiguous moderation rows, replace with clear-cut examples⁠ Running

    avanigupta +1: Claude Opus 4.6 (1M context)
    fcce834
  2. fix easy task test for updated issue types⁠

    avanigupta +1: Claude Opus 4.6 (1M context)
    0e13037
  3. replace ambiguous salary issue with date format fix⁠

    avanigupta +1: Claude Opus 4.6 (1M context)
    f1b7439
  4. fix root endpoint to list all 5 tasks⁠

    avanigupta +1: Claude Opus 4.6 (1M context)
    c3f32c9
  5. remove ambiguous LR fix — identify-only, any valid LR works⁠

    avanigupta +1: Claude Opus 4.6 (1M context)
    a1f98bf
  6. fix moderation issue row collisions and verify all data⁠

    avanigupta +1: Claude Opus 4.6 (1M context)
    8560706
  7. add moderation task to Gradio demo replay⁠

    avanigupta +1: Claude Opus 4.6 (1M context)
    887c1aa
  8. add content moderation task with real OpenAI Moderation data⁠

    avanigupta +1: Claude Opus 4.6 (1M context)
    b99e42b
  9. add toxic/biased response issue to alignment task⁠

    avanigupta +1: Claude Opus 4.6 (1M context)
    c699b6f
  10. replace ambiguous fixes with deterministic ones across all tasks⁠

    avanigupta +1: Claude Opus 4.6 (1M context)
    b08652c
  11. demo only proposes logically inferrable fixes⁠

    avanigupta +1: Claude Opus 4.6 (1M context)
    5de8f8e
  12. fix grading: reward valid fixes, not just exact matches⁠

    avanigupta +1: Claude Opus 4.6 (1M context)
    5e1f8bb
  13. update README with alignment task details and issue breakdown⁠

    avanigupta +1: Claude Opus 4.6 (1M context)
    1bd072d
  14. make alignment issues subtler to challenge frontier models⁠

    avanigupta +1: Claude Opus 4.6 (1M context)
    96d698c
  15. fix alignment demo trajectory to use correct clean values for fixes⁠

    avanigupta +1: Claude Opus 4.6 (1M context)
    8910a26
  16. use real NVIDIA HelpSteer data for alignment task⁠

    avanigupta +1: Claude Opus 4.6 (1M context)
    4051320
  17. improve alignment task: replace label swaps with real contamination⁠

    avanigupta +1: Claude Opus 4.6 (1M context)
    a9620ef
  18. use real Stanford Alpaca data for alignment task⁠

    avanigupta +1: Claude Opus 4.6 (1M context)
    7479de3
  19. add alignment data QA task: 12 issues in LLM instruction-tuning data⁠

    avanigupta +1: Claude Opus 4.6 (1M context)
    5cb467d
  20. Fix port to 8000 for validator compatibility⁠

    Varshith B +1: Claude Opus 4.6 (1M context)
    56f55e9
  21. Add root-level wrapper files and uv.lock for openenv deployment⁠

    Varshith B +1: Claude Opus 4.6 (1M context)
    0dbc19e
  22. Merge pull request #1 from varshith15/enhancementsv1⁠

    Varshith Bathini
    Unverified signature
    ca01572
  23. remove base_path: /web to fix HF Space iframe 404⁠

    avanigupta +1: Claude Opus 4.6 (1M context)
    85257bc
  24. add root endpoint for browser/judge friendliness⁠

    avanigupta +1: Claude Opus 4.6 (1M context)
    51adf89
  25. remove binary PNGs for HF push compatibility⁠

    avanigupta +1: Claude Opus 4.6 (1M context)
    d7c51ad
  26. use port 7860 for HF Spaces compatibility⁠

    avanigupta +1: Claude Opus 4.6 (1M context)
    671acb9
  27. minor change for meeting requirement format

    92187c5
  28. clean code structure

    22369d8
  29. expand datasets to include harder real-world scenarios

    5d90461
  30. expand datasets

    081eb22
  31. add fix stage+demo

    c3002ad
  32. fixes v1: add per step reward

    cd11aba

  1. init

    Varshith B
    4c1a85d