Hugging Face's logo

Spaces:
Tyr-123
/
TEST
Sleeping

App Files Community
TEST

Commit History

  1. Rename benchmarks/README.md to benchmarks/benchmarkinfo.md

    Tyr-123
    Verified signature
    92fd542

  2. Upload README.md

    Tyr-123
    Verified signature
    4e9f968

  3. Add GRPO evidence bundle⁠

    Tyr-123
    2a83133

  4. Upload README.md

    Tyr-123
    Verified signature
    52c38ac

  5. Upload BLOG.md

    Tyr-123
    Verified signature
    b372c10

  6. Clarify README training evidence

    Tyr-123
    52054e0

  7. Fix observation runtime fields

    Tyr-123
    5c561c3

  8. Update README for submission

    Tyr-123
    90c08ce

  9. Update graph images with latest visual revisions

    Tyr-123
    c5b4252

  10. Update README and final assets for OPENENV-PROJECT

    Tyr-123
    776ae8a

  11. Upload score_comparison.png with huggingface_hub

    Tyr-123
    Verified signature
    9cc5517

  12. Update generate_training_curves.py

    Tyr-123
    Verified signature
    d71f7cd

  1. Upload artifacts/training_results_graph.png with huggingface_hub

    Tyr-123
    Verified signature
    ca88a99

  2. fix: handle device_map=auto, gradient checkpointing, and eager attention for stable PEFT training

    Tyr-123
    eb2bbf1

  3. fix: use tokenize=False + explicit tokenizer call to fix BatchEncoding shape error

    Tyr-123
    d8977bd

  4. fix: replace Unsloth with standard PEFT+BitsAndBytes to fix dtype crash on A100

    Tyr-123
    e0d2d1c

  5. sync: rebuild train_colab.ipynb to mirror train_grpo.py with dtype fix and pre/post eval

    Tyr-123
    c2912e0

  6. fix: explicitly set model dtype to avoid float/half mismatch in unsloth

    Tyr-123
    430e097

  7. cleanup: remove unnecessary files (PDFs, duplicate PNGs, lock files, PDF generators)

    Tyr-123
    c9f8e3d

  8. audit: fix HF Space link, remove TLDB files, fix paths, update training refs to 7B

    Tyr-123
    64092fb

  9. cleanup: remove small test notebook, keep only 7B competition build

    Tyr-123
    a414c07

  10. fix: rewrite both notebooks with live logging, prompt key, bf16 fix

    Tyr-123
    ad912a0

  11. feat: clean live sample logging in exact user format

    Tyr-123
    9175a9f

  12. fix: change dataset key from 'messages' to 'prompt' for TRL GRPOTrainer

    Tyr-123
    d08a14d

  13. fix: auto rm -rf before git clone in colab notebooks

    Tyr-123
    ae7f601

  14. fix: unsloth batch math and bf16/fp16 dynamic assignment for T4/A100

    Tyr-123
    46f56e2

  15. fix: add openenv-core to pip install in both training notebooks

    Tyr-123
    e37d8f7

  16. fix: add training-safe SDK fallback + upgrade to 7B + test notebook

    Tyr-123
    ec07754

  17. fix: remove SDK ImportError fallback — fail fast with clear ModuleNotFoundError

    Tyr-123
    ea6f9f6

  1. Final bug fixes, dashboard io patch, and OpenEnv hackathon preparation

    Tyr-123
    36cc719

  1. Add LFS tracking for binary files (PNG, PDF, ipynb)

    Tyr-123
    6bb92ee

  2. SocialContract-v0: Complete hackathon submission with multi-agent, curriculum, GRPO training, benchmarks, and interactive demo

    Tyr-123
    8b275fc