Hugging Face's logo

Spaces:
agentDebugger
/
AgentDebugger-training-v3
Sleeping

App Files Community
AgentDebugger-training-v3

Commit History

  1. fix: resolve pip dependency conflicts for HF Spaces build⁠

    shank
    d0d5f60

  1. Fix: loosen strict dependencies to prevent pip backtracking

    shank
    2e3be87

  2. fix: remove hardcoded torch from requirements for HF space

    shank
    fe04772

  3. chore: normalize dataset inputs and fix mergekit dependency for TRL 0.14.0

    shank
    e67270e

  4. Add HANDOVER.md: full project state, deps, training instructions, known fixes⁠

    shank
    97aad17

  5. Auto-detect GPU: bfloat16+batch2+gen8 on A100, float16+batch1+gen4 on T4 — same script works on both⁠

    shank
    ea6fe4e

  6. Reduce max_completion_length to 160 for T4 speed: target 1000 steps in <8hrs⁠

    shank
    9487853

  7. Fix: bump bitsandbytes to 0.45.3 for CUDA 12.x support on Kaggle T4⁠

    shank
    6bf2fbb

  8. Optimize for Kaggle P100: float16, batch=1, grad_accum=8, num_gen=4, max_completion=256, lora_r=8⁠

    shank
    73f957d

  9. Fix GRPOConfig: rename max_new_tokens to max_completion_length for trl==0.14.0⁠

    shank
    8b16369

  10. Update: Added testing

    shank
    a5c67b3

  11. Align gradio version with Hugging Face Space builder2

    shank
    633a3b7

  12. Add dockerignore to reduce Space build context

    shank
    c945597

  13. Stabilize Space runtime: pin ML deps and disable runtime package drift

    shank
    663b8db

  14. Pin torch to cu121 build + use model.device instead of hardcoded cuda string⁠

    shank
    8f291e0

  15. Replace unsloth with bitsandbytes+peft: fixes CUDA driver incompatibility on HF A100⁠

    shank
    c325ad7

  16. Fix Gradio 4.x every= deprecation: use gr.Timer for auto-refresh⁠

    shank
    5eea2dd

  17. Reduce training to 500 steps with tightened curriculum for A10G budget⁠

    shank
    ba8df98

  18. Add Gradio training monitor and fix subprocess python path⁠

    shank
    b37b2eb

  19. Fix eval device selection with CUDA-safe fallback

    shank
    dc8001b

  20. Add interactive Gradio demo at /demo in env Space⁠

    shank
    4bac574

  21. Optimize for A100 80GB: 8 generations, batch 4, lr 2e-5, dense logging⁠

    shank
    2b1fbf3

  22. Restore full 1000-step training with original curriculum⁠

    shank
    1128de1

  23. Reduce training to 500 steps with tightened curriculum for A10G budget⁠

    shank
    3152fa9

  24. "Update: Minor fixes"

    shank
    755a07d

  25. Add Gradio training monitor and fix subprocess python path⁠

    shank
    b92ad01

  26. Curated the bugs dataset

    PulipatiPranav
    85f14d3

  27. Resolved README Merge conflicts

    PulipatiPranav
    a693c08

  28. Update: Started making changes for the hackathon

    shank
    a55c81d

  1. Updated README.md

    Pranav Pulipati
    a849e43

  1. Update: Refined and validated values

    shank
    6cca39d

  2. Update: Even more Final README.md update

    shank
    5c507c3

  3. Update: Final README.md update

    shank
    3548cd0

  4. Update: Final README.md update

    shank
    4057375

  5. Fix: Changed environment variables and added validator

    shank
    e93446d

  1. Fix: Final submission cleanup, unified identity and integrity markers

    shank
    8807d25

  2. Update: Made refinements to the project

    shank
    159a5fa

  3. Fix: Revise README for improved clarity and detail⁠

    Shashaank
    1c8aca2

  1. Update: README⁠

    Shashaank
    e4f09cc

  2. Fix: Changed scores range

    shank
    0769caa

  3. Fix: Changed dependencies

    shank
    ea7105c

  4. Fix: Change in gitignore

    shank
    cd3a400

  5. Fix: Remove gitignore from commits

    shank
    62a3a90

  6. Fix: Formatted the output

    shank
    0181886

  7. Update: Dockerfile and inference.py

    shank
    212d2d9

  8. Fix: precaution to prevent infinite loop

    shank
    cd968e7

  9. Fix: Fixed exception handling in inference.py

    shank
    2a482a5

  1. docs: updated the readme and added a license

    shank
    a2ff803

  2. docs: updated the readme and added a license

    shank
    22cb7e7

  3. fix: score floor for medium grader, add root and tasks endpoints

    shank
    b658e10