Commit History

  1. feat: run easy/medium/hard sequentially with START/END logs

    TanujInsane
    6e39a51

  1. fix: use task difficulty ID in [END] line for task validation

    TanujInsane
    cfd8bf6
  2. fix: LLM budget + keyword fallback + num_episodes=1 + closure bug fix

    TanujInsane
    6cfc309
  3. fix: use AgentGrader directly instead of HTTP calls

    TanujInsane
    e7e59dd
  4. fix: clip score to (0.05, 0.95) to avoid exact 0.0 or 1.0

    TanujInsane
    b06dcfb
  5. fix: track actual steps for accurate score calculation

    TanujInsane
    581ad49
  6. fix: run 3 difficulties with 10 steps each for task validation

    TanujInsane
    19d3a15
  7. fix: add LLM classification via API_BASE_URL and API_KEY

    TanujInsane
    e263f9c

  1. fix: add wait_for_server and remove sys.exit(1)

    TanujInsane
    676fd8d

  1. fix: use HF Space URL in inference.py, not localhost

    TanujInsane
    3a86e09
  2. fix: add required [START]/[STEP]/[END] structured output to inference.py

    TanujInsane
    860a4b8
  3. fix: add wait_for_server retry loop for phase 2

    TanujInsane
    53c6dcf
  4. Update inference.py retry loop for server connection

    TanujInsane
    1b64cdd

  1. fix: update inference.py with API_BASE_URL and HTTP client format

    TanujInsane
    99d66ea